Begin reimplementing compose

Re-theme
Implement Manticore fully and re-theme
2026-03-12 00:12:57 +00:00 · 2026-03-11 15:12:17 +00:00 · 2026-03-11 02:19:08 +00:00 · 2026-03-09 12:28:30 +00:00 · 2026-03-08 23:16:15 +00:00 · 2026-03-08 22:08:55 +00:00
395 changed files with 60556 additions and 6008 deletions
--- a/.dockerignore
+++ b/.dockerignore
@@ -0,0 +1,9 @@
+.git
+.podman
+artifacts
+.container-home
+db.sqlite3
+docker/data
+signal-cli-config
+__pycache__
+*.pyc
--- a/.gitignore
+++ b/.gitignore
@@ -166,3 +166,6 @@ oom
 node_modules/
 .podman/
 .beads/
+.sisyphus/
+
+.container-home/
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -1,22 +1,22 @@
 repos:
  - repo: https://github.com/psf/black
-    rev: 23.1.0
+    rev: 26.3.0
    hooks:
    - id: black
      exclude: ^core/migrations
  - repo: https://github.com/PyCQA/isort
-    rev: 5.11.5
+    rev: 8.0.1
    hooks:
      - id: isort
        args: ["--profile", "black"]
  - repo: https://github.com/PyCQA/flake8
-    rev: 6.0.0
+    rev: 7.3.0
    hooks:
      - id: flake8
        args: [--max-line-length=88]
        exclude: ^core/migrations
  - repo: https://github.com/rtts/djhtml
-    rev: v2.0.0
+    rev: 3.0.10
    hooks:
      - id: djhtml
        args: [-t 2]
@@ -25,6 +25,6 @@ repos:
      - id: djjs
        exclude: ^core/static/js # slow
  - repo: https://github.com/sirwart/ripsecrets.git
-    rev: v0.1.5
+    rev: v0.1.11
    hooks:
      - id: ripsecrets
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -2,7 +2,7 @@

 ## Overview

-GIA is a multi-transport communication platform bridging Signal, WhatsApp, XMPP, and Instagram through a Django web interface. It provides message relay, AI-powered workspace analysis, compose UX, and OSINT search. Stack: Python 3.11, Django 4.x, HTMX, Bulma CSS, SQLite, Redis, Docker Compose. Async runtime uses asyncio + uvloop.
+GIA is a multi-transport communication platform bridging Signal, WhatsApp, XMPP, and Instagram through a Django web interface. It provides message relay, AI-powered workspace analysis, compose UX, and OSINT search. Stack: Python 3.11, Django 4.x, HTMX, Bulma CSS, SQLite, Redis, Podman. Async runtime uses asyncio + uvloop.

 ## Structure

@@ -45,8 +45,8 @@ GIA/
 │   ├── templates/        # Django templates (75 files, partials/ heavy)
 │   ├── management/commands/ # ur (unified router), scheduling
 │   └── util/logs.py      # Custom colored logger — use logs.get_logger("name")
-├── Makefile              # Docker Compose orchestration commands
-├── docker-compose.yml    # Services: app, asgi, ur, scheduling, redis, signal-cli
+├── Makefile              # Podman + quadlet orchestration commands
+├── scripts/quadlet/      # Podman lifecycle scripts and unit rendering
 ├── Dockerfile            # Python 3.11, venv at /venv
 ├── requirements.txt      # Pinned deps (django, openai, neonize, slixmpp, etc.)
 ├── stack.env             # Runtime env vars (from stack.env.example)
@@ -56,14 +56,11 @@ GIA/
 ## Commands

 ```bash
-# All commands run via Docker Compose with stack.env
+# All commands run via Podman + quadlet with stack.env
 make build                          # Build Docker images
 make run                            # Start all services (quadlet manager)
 make stop                           # Stop all services
 make log                            # Tail logs
-make compose-run                    # Start via docker-compose directly
-make compose-stop                   # Stop via docker-compose
-make compose-log                    # Tail via docker-compose

 # Database
 make migrate                        # Run Django migrations
@@ -80,8 +77,8 @@ python manage.py test core.tests.test_foo.TestBar -v 2  # Single class
 python manage.py test core.tests.test_foo.TestBar.test_method -v 2  # Single test

 # Service restarts after code changes
-docker-compose restart ur           # Restart unified router
-docker-compose restart scheduling   # Restart scheduler
+podman restart ur_gia               # Restart unified router
+podman restart scheduling_gia       # Restart scheduler
 # uWSGI auto-reloads for app/core code changes
 ```

@@ -169,5 +166,5 @@ docker-compose restart scheduling   # Restart scheduler
 - **Unified Router** (`core/modules/router.py`): Management command `python manage.py ur` runs the event loop with all transport clients. Each client inherits `ClientBase` ABC.
 - **Transport Layer** (`core/clients/transport.py`): Shared cache-backed runtime state, command queuing, and attachment prep. All outbound media goes through `prepare_outbound_attachments()`.
 - **Settings Chain**: `app/settings.py` → imports `app/local_settings.py` (wildcard `*`) → env vars from `stack.env`. Feature flags: `WHATSAPP_ENABLED`, `INSTAGRAM_ENABLED`, `COMPOSE_WS_ENABLED`.
- **Services in docker-compose**: `app` (uWSGI), `asgi` (uvicorn for WebSockets), `ur` (unified router), `scheduling` (APScheduler), `redis`, `signal-cli-rest-api`.
+- **Services in podman stack**: `app` (uWSGI), `asgi` (uvicorn for WebSockets), `ur` (unified router), `scheduling` (APScheduler), `redis`, `signal-cli-rest-api`.
 - **No test suite currently**: `core/tests.py` is empty scaffold; `core/tests/` has only `__init__.py`. Tests run via `make test MODULES=...` but need to be written.
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -0,0 +1,61 @@
+# GIA — Claude Code Rules
+
+## Privacy: No Real Contact Data in Code
+
+**NEVER use real contact identifiers in tests, fixtures, seeds, or any committed file.**
+
+Real contact data includes: phone numbers, JIDs, email addresses, usernames, or any identifier belonging to an actual person in the user's contacts.
+
+### Use fictitious data instead
+
+| Type | Safe fictitious examples |
+|---|---|
+| UK mobile (E.164) | `+447700900001`, `+447700900002` (Ofcom-reserved range 07700 900000–900999) |
+| UK mobile (no +) | `447700900001`, `447700900002` |
+| US phone | `+15550001234`, `+15550009999` (555-0xxx NANP reserved range) |
+| Email | `test@example.com`, `user@example.invalid` |
+| WhatsApp JID | `447700900001@s.whatsapp.net`, `447700900001@g.us` |
+
+### Why this matters
+
+AI coding tools (Copilot, Claude) will reuse any values they see in context. A real number placed in a test becomes training signal and will be suggested in future completions — potentially leaking it further.
+
+### Quick check
+
+Before committing test files, verify no identifier matches a real person:
+- No number outside the reserved fictitious ranges above
+- No name that corresponds to a real contact used as a literal identifier
+
+## Naming: Avoid Ambiguous Role Labels
+
+**Never use "User", "Bot", "Us", or "Them" as role labels without qualification — these terms are context-dependent and misleading in this codebase.**
+
+GIA acts in multiple roles simultaneously:
+- It is a Django **User** (account holder) from the perspective of external services (XMPP, WhatsApp, Signal).
+- It is a **component** (gateway/bot) from the perspective of contacts.
+- The human who owns and operates the GIA instance is the **account holder** or **operator** (not "user", which collides with `User` model).
+- Remote people the system communicates with are **contacts**.
+
+Preferred terms:
+
+| Avoid              | Prefer                                                          |
+| ------------------ | --------------------------------------------------------------- |
+| "User" (ambiguous) | "account holder" or "operator" (for the Django `User`)         |
+| "Bot"              | "component" or "gateway" (for the XMPP/transport layer)        |
+| "Us"               | name the specific actor: "GIA", "the component", "the operator" |
+| "Them"             | "contact" or "remote party"                                     |
+
+Apply this in: comments, template labels, log messages, and variable names.
+
+## Runtime: uWSGI Reload File
+
+The app container uses uWSGI with a single touch-based reload sentinel:
+
+- Reload file: `/code/.uwsgi-reload`
+- Config: [docker/uwsgi.ini](/code/xf/GIA/docker/uwsgi.ini)
+
+Rules:
+
+- Never run `python manage.py ...` on the host. Run Django management commands inside Podman, for example with `podman exec`.
+- After changing templates or app code that should be picked up by the `gia` uWSGI service, touch `/code/.uwsgi-reload`.
+- If the uWSGI config itself changes, touch `/code/.uwsgi-reload` and restart the `gia` container so the new config is loaded.
--- a/Containerfile.dev
+++ b/Containerfile.dev
@@ -0,0 +1,56 @@
+FROM python:3.11-bookworm
+
+ARG USER_ID=1000
+ARG GROUP_ID=1000
+ARG USER_NAME=dev
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    bash-completion \
+    build-essential \
+    cargo \
+    ca-certificates \
+    curl \
+    fd-find \
+    fzf \
+    git \
+    golang \
+    jq \
+    less \
+    libffi-dev \
+    libssl-dev \
+    make \
+    neovim \
+    nodejs \
+    npm \
+    procps \
+    ripgrep \
+    rsync \
+    rustc \
+    sqlite3 \
+    tar \
+    tmux \
+    unzip \
+    wget \
+    which \
+    zip && \
+    rm -rf /var/lib/apt/lists/* && \
+    ln -sf /usr/bin/fdfind /usr/local/bin/fd
+
+RUN groupadd -g "${GROUP_ID}" "${USER_NAME}" 2>/dev/null || true && \
+    useradd -m -u "${USER_ID}" -g "${GROUP_ID}" -s /bin/bash "${USER_NAME}"
+
+USER ${USER_NAME}
+WORKDIR /home/${USER_NAME}
+
+# Build a project virtualenv and preinstall dependencies.
+COPY --chown=${USER_NAME}:${USER_NAME} requirements.txt /tmp/requirements.txt
+RUN bash -lc 'set -e; \
+    python3.11 -m venv /home/${USER_NAME}/.venv/gia && \
+    /home/${USER_NAME}/.venv/gia/bin/pip install --upgrade pip setuptools wheel && \
+    grep -Ev "^(git\\+https://git\\.zm\\.is/|aiograpi$)" /tmp/requirements.txt > /tmp/requirements.build.txt && \
+    /home/${USER_NAME}/.venv/gia/bin/pip install -r /tmp/requirements.build.txt'
+
+ENV VIRTUAL_ENV=/home/${USER_NAME}/.venv/gia
+ENV PATH="${VIRTUAL_ENV}/bin:${PATH}"
+
+CMD ["/bin/bash"]
--- a/8
+++ b/8
@@ -14,17 +14,13 @@ RUN chown xf:xf /venv

 RUN apt-get update && apt-get install -y cargo rustc

-USER xf
 ENV PYTHONDONTWRITEBYTECODE=1
 ENV PYTHONUNBUFFERED=1
 WORKDIR /code
 COPY requirements.txt /code/
 RUN python -m venv /venv
 RUN . /venv/bin/activate && pip install -r requirements.txt
+RUN chown -R xf:xf /code /venv /conf

-# CMD . /venv/bin/activate && uwsgi --ini /conf/uwsgi.ini
-
+USER xf
 CMD if [ "$OPERATION" = "uwsgi" ] ; then . /venv/bin/activate && uwsgi --ini /conf/uwsgi.ini ; else . /venv/bin/activate && exec python manage.py runserver 0.0.0.0:8000; fi
-
-# CMD . /venv/bin/activate && uvicorn --reload --reload-include *.html --workers 2 --uds /var/run/socks/app.sock app.asgi:application
-# CMD . /venv/bin/activate && gunicorn -b 0.0.0.0:8000 --reload app.asgi:application -k uvicorn.workers.UvicornWorker
--- a/INSTALL.md
+++ b/INSTALL.md
@@ -6,7 +6,7 @@ Use this first. Then read `README.md` for feature and operation-mode details.

 ## 1) Prerequisites

- Linux host with either Podman + podman-compose wrapper or Docker Compose compatibility.
+- Linux host with Podman.
 - Git.
 - Network access for service images and Python dependencies.

@@ -47,6 +47,24 @@ XMPP bridge settings:
 - `XMPP_PORT`
 - `XMPP_SECRET`

+Prosody container helpers:
+
+- `QUADLET_PROSODY_CONFIG_FILE`
+- `QUADLET_PROSODY_CERTS_DIR`
+- `QUADLET_PROSODY_DATA_DIR`
+- `QUADLET_PROSODY_LOGS_DIR`
+
+Memory/wiki search helpers:
+
+- `MEMORY_SEARCH_BACKEND` (`django` or `manticore`)
+- `MANTICORE_HTTP_URL`
+- `MANTICORE_MEMORY_TABLE`
+- `MANTICORE_EVENT_TABLE`
+- `MANTICORE_METRIC_TABLE`
+- `COMPOSING_ABANDONED_WINDOW_SECONDS`
+- `CONVERSATION_EVENT_RETENTION_DAYS`
+- `MANTICORE_HTTP_TIMEOUT`
+
 For XMPP media upload, configure one of:

 - `XMPP_UPLOAD_SERVICE`
@@ -71,9 +89,18 @@ make auth
 Optional static token helper:

 ```bash
-make token
+make token TOKEN_USER=<your_username>
 ```

+Local code-quality checks:
+
+```bash
+make pre-commit
+make pre-commit-glibc
+```
+
+`make pre-commit-glibc` selects `env` on musl systems and `genv` on glibc systems.
+
 ## 6) Logs and health checks

 Tail logs:
@@ -85,7 +112,7 @@ make log
 Basic stack status:

 ```bash
-docker-compose --env-file=stack.env ps
+make status
 ```

 ## 7) Restart conventions
@@ -101,7 +128,7 @@ Use the explicit `make stop && make run` command sequence when a full recycle is
 ### Single service restart

 ```bash
-docker-compose --env-file=stack.env restart <service>
+podman restart <container-name>
 ```

 If single-service restart fails due to dependency/pod state, use full recycle above.
@@ -113,7 +140,7 @@ After changing UR/runtime code (`core/clients/*`, transport, relay paths), resta
 Minimum target:

 ```bash
-docker-compose --env-file=stack.env restart ur
+podman restart ur_gia
 ```

 If blocked, use full recycle.
@@ -127,8 +154,21 @@ If blocked, use full recycle.
 - Manual compose: `/compose/page/`
 - AI workspace: `/ai/workspace/`
 - OSINT search: `/search/page/`
+- Security encryption settings: `/settings/security/encryption/`
+- Security permissions settings: `/settings/security/permissions/`
+- Command routing settings: `/settings/command-routing/`
+- Task automation settings: `/settings/tasks/`
+- Task inbox / manual task creation: `/tasks/`

-## 10) Common troubleshooting
+## 10) Security and capability controls
+
+- `Require OMEMO encryption` rejects plaintext XMPP messages before command routing.
+- `Encrypt gateway component chat replies with OMEMO` only affects gateway/component conversations.
+- `Encrypt contact relay messages to your XMPP client with OMEMO` only affects relayed contact chats.
+- Fine-grained capability policy is configured in `/settings/security/permissions/` and applies by scope, service, and optional channel pattern.
+- Trusted OMEMO key enforcement depends on trusted key records, not only the most recently observed client key.
+
+## 11) Common troubleshooting

 ### A) Compose restart errors / dependency improper state

@@ -144,8 +184,131 @@ make stop && make run
 - Confirm `XMPP_UPLOAD_SERVICE`/`XMPP_UPLOAD_JID` is set, or discovery works.
 - Check runtime logs for slot request and upload errors.

+### D) Prosody container (lightweight, no systemd)
+
+Use:
+
+```bash
+./utilities/prosody/manage_prosody_container.sh up
+./utilities/prosody/manage_prosody_container.sh status
+./utilities/prosody/manage_prosody_container.sh logs
+```
+
+Auth script path for Prosody config:
+
+```lua
+external_auth_command = "/code/utilities/prosody/auth_django.sh"
+```
+
+Certificate renewal helper (run as root on host):
+
+```bash
+./utilities/prosody/renew_prosody_cert.sh
+```
+
+### E) Manticore container for memory/wiki retrieval
+
+```bash
+./utilities/memory/manage_manticore_container.sh up
+./utilities/memory/manage_manticore_container.sh status
+./utilities/memory/manage_manticore_container.sh logs
+```
+
+Reindex memory into configured backend:
+
+```bash
+podman exec ur_gia /venv/bin/python manage.py memory_search_reindex --user-id 1 --statuses active
+```
+
+Query memory backend:
+
+```bash
+podman exec ur_gia /venv/bin/python manage.py memory_search_query --user-id 1 --query "reply style"
+```
+
+Generate proposed memories from recent inbound messages:
+
+```bash
+podman exec ur_gia /venv/bin/python manage.py memory_suggest_from_messages --user-id 1 --limit-messages 300 --max-items 30
+```
+
+Run memory hygiene (expiry decay + contradiction queueing):
+
+```bash
+podman exec ur_gia /venv/bin/python manage.py memory_hygiene --user-id 1
+```
+
+Performance defaults now applied in GIA:
+
+- Batched Manticore reindex writes (`REPLACE ... VALUES (...)` in chunks) for lower ingest latency.
+- Cached table-ensure checks to avoid `CREATE TABLE IF NOT EXISTS` overhead on every query.
+- Behavioral event dual-write uses `MANTICORE_EVENT_TABLE` (default `gia_events`) when event ledger flags are enabled.
+- Behavioral metrics are written by `python manage.py gia_analysis` into `MANTICORE_METRIC_TABLE` (default `gia_metrics`).
+- ORM shadow copies can be pruned with `python manage.py prune_behavioral_orm_data`; defaults are driven by `CONVERSATION_EVENT_RETENTION_DAYS`.
+- Runtime table maintenance available through MCP (`FLUSH RAMCHUNK`, `OPTIMIZE TABLE`) for steady query responsiveness.
+
+### F) MCP server for task + memory tooling (VS Code)
+
+The workspace includes an MCP config at `/code/xf/.vscode/mcp.json` for server `manticore`.
+
+It launches inside the running `ur_gia` container and forces:
+
+- `MEMORY_SEARCH_BACKEND=manticore`
+
+`MANTICORE_HTTP_URL` is read from the container environment (`stack.env` / app settings).
+
+Start requirements first:
+
+```bash
+make run
+./utilities/memory/manage_manticore_container.sh up
+```
+
+Then approve/enable the `manticore` MCP server in VS Code when prompted.
+
+The MCP task surface now supports canonical task creation/completion in GIA:
+
+- `tasks.create`
+- `tasks.complete`
+- `tasks.create_note`
+- `tasks.link_artifact`
+
+Optional ultra-light Rust MCP worker:
+
+```bash
+cd /code/xf/GIA
+make mcp-rust-build
+```
+
+Then enable `manticore-rust-worker` in `/code/xf/.vscode/mcp.json`.
+It is intentionally `disabled: true` by default so the existing Python MCP server remains the baseline.
+
+### H) Optional browser MCP for visual validation
+
+To validate compose/tasks/settings flows visually, add a browser-capable MCP server in your editor workspace alongside `manticore`. A Playwright-style browser MCP is the intended integration point for GIA UI checks.
+
+Recommended usage:
+
+- keep browser MCP outside host-network mode
+- point it at the local GIA app URL/port from the running stack
+- use it for page-load, form-flow, and visual regression checks on compose/tasks/settings pages
+
+### I) Task command shortcuts
+
+Gateway / XMPP / chat task commands now include:
+
+- `.l` -> list open tasks
+- `.tasks add <project> :: <title>` -> create a canonical task in a named project
+- `.task add <title>` -> create a task inside the current mapped chat scope
+
 ### C) Signal or WhatsApp send failures

 - Verify account/link status in service pages.
 - Verify `ur` service is running.
 - Inspect `ur` logs for transport-specific errors.
+
+### G) XMPP reconnect loop in logs
+
+- Confirm `XMPP_ADDRESS`, `XMPP_JID`, `XMPP_PORT`, and `XMPP_SECRET` are populated in `stack.env`.
+- `XMPP_PORT` is parsed as an integer in settings; invalid values can cause repeated reconnect failures.
+- The runtime now uses a single reconnect loop with exponential backoff to avoid overlapping reconnect churn.
--- a/74
+++ b/74
@@ -1,10 +1,18 @@
 QUADLET_MGR := ./scripts/quadlet/manage.sh
+MODULES ?= core.tests
+TOKEN_USER ?= m
+STACK_ID_CLEAN := $(shell sid="$${GIA_STACK_ID:-$${STACK_ID:-}}"; sid=$$(printf "%s" "$$sid" | tr -cs 'a-zA-Z0-9._-' '-' | sed 's/^-*//; s/-*$$//'); printf "%s" "$$sid")
+STACK_SUFFIX := $(if $(STACK_ID_CLEAN),_$(STACK_ID_CLEAN),)
+APP_CONTAINER := gia$(STACK_SUFFIX)
+LOCAL_LIBC := $(shell if ldd --version 2>&1 | head -n1 | tr '[:upper:]' '[:lower:]' | grep -q musl; then printf musl; else printf glibc; fi)
+LOCAL_VENV := $(if $(filter musl,$(LOCAL_LIBC)),env,genv)
+PRE_COMMIT_BIN := $(firstword $(wildcard $(LOCAL_VENV)/bin/pre-commit) $(wildcard genv/bin/pre-commit) $(wildcard env/bin/pre-commit))

 run:
 	bash $(QUADLET_MGR) up

 build:
-	docker-compose --env-file=stack.env build
+	OPERATION=uwsgi podman build --build-arg OPERATION=uwsgi -t localhost/xf/gia:prod -f Dockerfile .

 stop:
 	bash $(QUADLET_MGR) down
@@ -18,26 +26,62 @@ status:
 quadlet-install:
 	bash $(QUADLET_MGR) install

-compose-run:
-	docker-compose --env-file=stack.env up -d
-
-compose-stop:
-	docker-compose --env-file=stack.env down
-
-compose-log:
-	docker-compose --env-file=stack.env logs -f --names
-
 test:
-	docker-compose --env-file=stack.env run --rm app sh -c ". /venv/bin/activate && python manage.py test $(MODULES) -v 2"
+	@if podman ps --format '{{.Names}}' | grep -qx "$(APP_CONTAINER)"; then \
+		podman exec "$(APP_CONTAINER)" sh -lc "cd /code && . /venv/bin/activate && python manage.py test $(MODULES) -v 2"; \
+	else \
+		echo "Container '$(APP_CONTAINER)' is not running. Start the stack first with 'make run'." >&2; \
+		exit 125; \
+	fi
+
+pre-commit:
+	@if [ -x "$(PRE_COMMIT_BIN)" ]; then \
+		"$(PRE_COMMIT_BIN)" run -a; \
+	else \
+		echo "No local pre-commit executable found in $(LOCAL_VENV)/bin, genv/bin, or env/bin." >&2; \
+		exit 127; \
+	fi
+
+pre-commit-glibc:
+	@if [ -x "$(PRE_COMMIT_BIN)" ]; then \
+		echo "Using $(LOCAL_VENV) ($(LOCAL_LIBC))"; \
+		"$(PRE_COMMIT_BIN)" run -a; \
+	else \
+		echo "No local pre-commit executable found in $(LOCAL_VENV)/bin, genv/bin, or env/bin." >&2; \
+		exit 127; \
+	fi

 migrate:
-	docker-compose --env-file=stack.env run --rm app sh -c ". /venv/bin/activate && python manage.py migrate"
+	@if podman ps --format '{{.Names}}' | grep -qx "$(APP_CONTAINER)"; then \
+		podman exec "$(APP_CONTAINER)" sh -lc "cd /code && . /venv/bin/activate && python manage.py migrate"; \
+	else \
+		echo "Container '$(APP_CONTAINER)' is not running. Start the stack first with 'make run'." >&2; \
+		exit 125; \
+	fi

 makemigrations:
-	docker-compose --env-file=stack.env run --rm app sh -c ". /venv/bin/activate && python manage.py makemigrations"
+	@if podman ps --format '{{.Names}}' | grep -qx "$(APP_CONTAINER)"; then \
+		podman exec "$(APP_CONTAINER)" sh -lc "cd /code && . /venv/bin/activate && python manage.py makemigrations"; \
+	else \
+		echo "Container '$(APP_CONTAINER)' is not running. Start the stack first with 'make run'." >&2; \
+		exit 125; \
+	fi

 auth:
-	docker-compose --env-file=stack.env run --rm app sh -c ". /venv/bin/activate && python manage.py createsuperuser"
+	@if podman ps --format '{{.Names}}' | grep -qx "$(APP_CONTAINER)"; then \
+		podman exec "$(APP_CONTAINER)" sh -lc "cd /code && . /venv/bin/activate && python manage.py createsuperuser"; \
+	else \
+		echo "Container '$(APP_CONTAINER)' is not running. Start the stack first with 'make run'." >&2; \
+		exit 125; \
+	fi

 token:
-	docker-compose --env-file=stack.env run --rm app sh -c ". /venv/bin/activate && python manage.py addstatictoken m"
+	@if podman ps --format '{{.Names}}' | grep -qx "$(APP_CONTAINER)"; then \
+		podman exec "$(APP_CONTAINER)" sh -lc "cd /code && . /venv/bin/activate && python manage.py addstatictoken $(TOKEN_USER)"; \
+	else \
+		echo "Container '$(APP_CONTAINER)' is not running. Start the stack first with 'make run'." >&2; \
+		exit 125; \
+	fi
+
+mcp-rust-build:
+	cd rust/manticore-mcp-worker && cargo build --release
--- a/README.md
+++ b/README.md
@@ -9,9 +9,15 @@ GIA is a multi-transport communication workspace that unifies Signal, WhatsApp,
 - Unifies chats from multiple protocols in one interface.
 - Keeps conversation history in a shared model (`Person`, `PersonIdentifier`, `ChatSession`, `Message`).
 - Supports manual, queue-driven, and AI-assisted outbound messaging.
+- Supports canonical task creation from chat commands, web UI, and MCP tooling.
 - Bridges messages across transports (including XMPP) with attachment handling.
 - Tracks delivery/read metadata and typing state events.
+- Can dual-write canonical behavioral events to Manticore for time-series analysis.
+- Includes `gia_analysis` for rolling behavioral metric aggregation into Manticore.
+- Includes `prune_behavioral_orm_data` to keep Django event shadow tables bounded once Manticore is primary.
 - Provides AI workspace analytics, mitigation plans, and insight visualizations.
+- Exposes fine-grained capability policy controls for gateway commands, task intake, and command execution.
+- Separates XMPP encryption controls into plaintext rejection, component-chat encryption, and relayed-contact encryption.

 ## Operation Modes

@@ -104,6 +110,14 @@ Core behavior:
 - XMPP bridge supports text, attachments, typing, and chat-state paths.
 - Signal and WhatsApp media relay paths are normalized via shared transport/media logic.

+## Settings Model
+
+- `Security > Encryption`: transport-level XMPP/OMEMO controls, observed client state, and discovered key trust management.
+- `Security > Permissions`: fine-grained capability policy by scope, service, and channel pattern.
+- `Modules > Commands`: command profiles, bindings, and delivery behavior.
+- `Modules > Task Automation`: task extraction defaults, channel overrides, and provider approval routing.
+- `Modules > Business Plans`: generated document inbox and editor.
+
 Key design points:

 - Prefer shared media preparation over per-service duplicated logic.
@@ -119,7 +133,9 @@ Core components:
 - `core/clients/signal.py`, `core/clients/signalapi.py`: Signal event + REST transport handling.
 - `core/clients/whatsapp.py`: Neonize-backed runtime transport.
 - `core/clients/xmpp.py`: XMPP component bridge and media upload relay.
+- `rust/manticore-mcp-worker`: optional ultra-light MCP frontend for direct Manticore status/query/maintenance.
 - `core/views/compose.py`: Manual compose UX, polling/ws, send pipeline, media blob endpoint.
+- `core/tasks/engine.py`: Canonical task creation/completion helpers used by chat commands and UI.
 - `core/views/workspace.py`: AI workspace operations and insight surfaces.
 - `core/views/osint.py`: Search/workspace OSINT interactions.

@@ -143,6 +159,7 @@ After environment setup from `INSTALL.md`:
 4. Open manual compose and test per-service send/receive.
 5. Open AI workspace for analysis/mitigation workflows.
 6. Verify queue workflows if approval mode is used.
+7. Verify task creation from `/tasks/`, `.tasks add <project> :: <title>`, or scoped `.task add <title>`.

 Recommended functional smoke test:

@@ -156,6 +173,7 @@ Recommended functional smoke test:
 - After runtime code changes, restart runtime services before validation.
 - Full environment recycle convention: `make stop && make run`.
 - If single-service restart fails due to dependency state, use full recycle.
+- Local repository checks are available via `make pre-commit`; use `make pre-commit-glibc` when you want libc-based `env`/`genv` selection.

 ## Security & Reliability Notes

--- a/app/local_settings.py
+++ b/app/local_settings.py
@@ -1,16 +1,39 @@
 from os import getenv
+from urllib.parse import urlparse

 trues = ("t", "true", "yes", "y", "1")

+
+def _csv_env(name: str, default: str) -> list[str]:
+    return [item.strip() for item in getenv(name, default).split(",") if item.strip()]
+
+
 # URLs
 DOMAIN = getenv("DOMAIN", "example.com")
 URL = getenv("URL", f"https://{DOMAIN}")
+URL_HOST = urlparse(URL).hostname or ""
+DEBUG = getenv("DEBUG", "false").lower() in trues

 # Access control
-ALLOWED_HOSTS = getenv("ALLOWED_HOSTS", f"127.0.0.1,{DOMAIN}").split(",")
+ALLOWED_HOSTS = _csv_env(
+    "ALLOWED_HOSTS",
+    ",".join(
+        item
+        for item in (
+            "localhost",
+            "127.0.0.1",
+            DOMAIN,
+            URL_HOST,
+        )
+        if item
+    ),
+)
+if DEBUG:
+    # Local/dev stack runs behind varying hostnames/tunnels.
+    ALLOWED_HOSTS = ["*"]

 # CSRF
-CSRF_TRUSTED_ORIGINS = getenv("CSRF_TRUSTED_ORIGINS", URL).split(",")
+CSRF_TRUSTED_ORIGINS = _csv_env("CSRF_TRUSTED_ORIGINS", URL)

 # Stripe
 BILLING_ENABLED = getenv("BILLING_ENABLED", "false").lower() in trues
@@ -23,7 +46,10 @@ STRIPE_PUBLIC_API_KEY_PROD = getenv("STRIPE_PUBLIC_API_KEY_PROD", "")

 STRIPE_ENDPOINT_SECRET = getenv("STRIPE_ENDPOINT_SECRET", "")
 STATIC_ROOT = getenv("STATIC_ROOT", "")
-SECRET_KEY = getenv("SECRET_KEY", "")
+SECRET_KEY = (getenv("SECRET_KEY", "") or "").strip()
+if not SECRET_KEY:
+    # Keep local developer stacks usable when stack.env is uninitialized.
+    SECRET_KEY = "gia-dev-secret-key"

 STRIPE_ADMIN_COUPON = getenv("STRIPE_ADMIN_COUPON", "")

@@ -33,17 +59,20 @@ LAGO_API_KEY = getenv("LAGO_API_KEY", "")
 LAGO_ORG_ID = getenv("LAGO_ORG_ID", "")
 LAGO_URL = getenv("LAGO_URL", "")

-DEBUG = getenv("DEBUG", "false") in trues
 PROFILER = getenv("PROFILER", "false") in trues

 if DEBUG:
    import socket  # only if you haven't already imported this

    hostname, _, ips = socket.gethostbyname_ex(socket.gethostname())
-    INTERNAL_IPS = [ip[: ip.rfind(".")] + ".1" for ip in ips] + [
-        "127.0.0.1",
-        "10.0.2.2",
-    ]
+    INTERNAL_IPS = [ip[: ip.rfind(".")] + ".1" for ip in ips]
+    INTERNAL_IPS.extend(
+        [
+            item.strip()
+            for item in getenv("DEBUG_INTERNAL_IPS", "localhost").split(",")
+            if item.strip()
+        ]
+    )

 SETTINGS_EXPORT = ["BILLING_ENABLED"]

@@ -58,5 +87,42 @@ INSTAGRAM_HTTP_URL = getenv("INSTAGRAM_HTTP_URL", "http://instagram:8080")

 XMPP_ADDRESS = getenv("XMPP_ADDRESS")
 XMPP_JID = getenv("XMPP_JID")
-XMPP_PORT = getenv("XMPP_PORT")
+XMPP_USER_DOMAIN = getenv("XMPP_USER_DOMAIN", "")
+XMPP_PORT = int(getenv("XMPP_PORT", "8888") or 8888)
 XMPP_SECRET = getenv("XMPP_SECRET")
+XMPP_OMEMO_DATA_DIR = getenv("XMPP_OMEMO_DATA_DIR", "")
+XMPP_UPLOAD_SERVICE = getenv("XMPP_UPLOAD_SERVICE", "").strip()
+XMPP_UPLOAD_JID = getenv("XMPP_UPLOAD_JID", "").strip()
+if not XMPP_UPLOAD_SERVICE and XMPP_UPLOAD_JID:
+    XMPP_UPLOAD_SERVICE = XMPP_UPLOAD_JID
+if not XMPP_UPLOAD_SERVICE and XMPP_USER_DOMAIN:
+    XMPP_UPLOAD_SERVICE = XMPP_USER_DOMAIN
+
+EVENT_LEDGER_DUAL_WRITE = getenv("EVENT_LEDGER_DUAL_WRITE", "false").lower() in trues
+CAPABILITY_ENFORCEMENT_ENABLED = (
+    getenv("CAPABILITY_ENFORCEMENT_ENABLED", "true").lower() in trues
+)
+TRACE_PROPAGATION_ENABLED = getenv("TRACE_PROPAGATION_ENABLED", "true").lower() in trues
+EVENT_PRIMARY_WRITE_PATH = getenv("EVENT_PRIMARY_WRITE_PATH", "false").lower() in trues
+
+MEMORY_SEARCH_BACKEND = getenv("MEMORY_SEARCH_BACKEND", "django")
+MANTICORE_HTTP_URL = getenv("MANTICORE_HTTP_URL", "http://localhost:9308")
+MANTICORE_MEMORY_TABLE = getenv("MANTICORE_MEMORY_TABLE", "gia_memory_items")
+MANTICORE_EVENT_TABLE = getenv("MANTICORE_EVENT_TABLE", "gia_events")
+MANTICORE_METRIC_TABLE = getenv("MANTICORE_METRIC_TABLE", "gia_metrics")
+COMPOSING_ABANDONED_WINDOW_SECONDS = int(
+    getenv("COMPOSING_ABANDONED_WINDOW_SECONDS", "300")
+)
+CONVERSATION_EVENT_RETENTION_DAYS = int(
+    getenv("CONVERSATION_EVENT_RETENTION_DAYS", "90") or 90
+)
+MANTICORE_HTTP_TIMEOUT = int(getenv("MANTICORE_HTTP_TIMEOUT", "5") or 5)
+
+# Attachment security defaults for transport adapters.
+ATTACHMENT_MAX_BYTES = int(getenv("ATTACHMENT_MAX_BYTES", str(25 * 1024 * 1024)) or 0)
+ATTACHMENT_ALLOW_PRIVATE_URLS = (
+    getenv("ATTACHMENT_ALLOW_PRIVATE_URLS", "false").lower() in trues
+)
+ATTACHMENT_ALLOW_UNKNOWN_MIME = (
+    getenv("ATTACHMENT_ALLOW_UNKNOWN_MIME", "false").lower() in trues
+)
--- a/app/settings.py
+++ b/app/settings.py
@@ -13,6 +13,8 @@ https://docs.djangoproject.com/en/4.0/ref/settings/
 import os
 from pathlib import Path

+TRUE_VALUES = {"1", "true", "yes", "on"}
+
 # Build paths inside the project like this: BASE_DIR / 'subdir'.
 BASE_DIR = Path(__file__).resolve().parent.parent

@@ -36,8 +38,6 @@ INSTALLED_APPS = [
    "django.contrib.sessions",
    "django.contrib.messages",
    "django.contrib.staticfiles",
-    "debug_toolbar",
-    "template_profiler_panel",
    "django_htmx",
    "crispy_forms",
    "crispy_bulma",
@@ -57,6 +57,16 @@ INSTALLED_APPS = [
    "cachalot",
 ]

+DEBUG_TOOLBAR_ENABLED = os.getenv("DEBUG_TOOLBAR_ENABLED", "false").lower() in TRUE_VALUES
+
+if DEBUG_TOOLBAR_ENABLED:
+    INSTALLED_APPS.extend(
+        [
+            "debug_toolbar",
+            "template_profiler_panel",
+        ]
+    )
+
 # Performance optimisations
 CACHES = {
    "default": {
@@ -81,7 +91,6 @@ CRISPY_ALLOWED_TEMPLATE_PACKS = ("bulma",)
 DJANGO_TABLES2_TEMPLATE = "django-tables2/bulma.html"

 MIDDLEWARE = [
-    "debug_toolbar.middleware.DebugToolbarMiddleware",
    "django.middleware.security.SecurityMiddleware",
    "django.contrib.sessions.middleware.SessionMiddleware",
    # 'django.middleware.cache.UpdateCacheMiddleware',
@@ -95,6 +104,9 @@ MIDDLEWARE = [
    "django_htmx.middleware.HtmxMiddleware",
 ]

+if DEBUG_TOOLBAR_ENABLED:
+    MIDDLEWARE.insert(0, "debug_toolbar.middleware.DebugToolbarMiddleware")
+
 ROOT_URLCONF = "app.urls"
 ASGI_APPLICATION = "app.asgi.application"
 COMPOSE_WS_ENABLED = os.environ.get("COMPOSE_WS_ENABLED", "false").lower() in {
@@ -104,19 +116,24 @@ COMPOSE_WS_ENABLED = os.environ.get("COMPOSE_WS_ENABLED", "false").lower() in {
    "on",
 }

+TEMPLATE_CONTEXT_PROCESSORS = [
+    "django.template.context_processors.request",
+    "django.contrib.auth.context_processors.auth",
+    "django.contrib.messages.context_processors.messages",
+    "core.util.django_settings_export.settings_export",
+    "core.context_processors.settings_hierarchy_nav",
+]
+
+if DEBUG_TOOLBAR_ENABLED:
+    TEMPLATE_CONTEXT_PROCESSORS.insert(0, "django.template.context_processors.debug")
+
 TEMPLATES = [
    {
        "BACKEND": "django.template.backends.django.DjangoTemplates",
        "DIRS": [os.path.join(BASE_DIR, "core/templates")],
        "APP_DIRS": True,
        "OPTIONS": {
-            "context_processors": [
-                "django.template.context_processors.debug",
-                "django.template.context_processors.request",
-                "django.contrib.auth.context_processors.auth",
-                "django.contrib.messages.context_processors.messages",
-                "core.util.django_settings_export.settings_export",
-            ],
+            "context_processors": TEMPLATE_CONTEXT_PROCESSORS,
        },
    },
 ]
@@ -130,7 +147,7 @@ WSGI_APPLICATION = "app.wsgi.application"
 DATABASES = {
    "default": {
        "ENGINE": "django.db.backends.sqlite3",
-        "NAME": "/conf/db.sqlite3",
+        "NAME": os.getenv("APP_DATABASE_PATH", "/conf/db.sqlite3"),
    }
 }

@@ -189,8 +206,9 @@ REST_FRAMEWORK = {
 }

 INTERNAL_IPS = [
-    "127.0.0.1",
-    "10.1.10.11",
+    item.strip()
+    for item in os.getenv("INTERNAL_IPS", "localhost").split(",")
+    if item.strip()
 ]

 DEBUG_TOOLBAR_PANELS = [
@@ -228,7 +246,7 @@ if PROFILER:  # noqa - trust me its there


 def show_toolbar(request):
-    return DEBUG  # noqa: from local imports
+    return DEBUG and DEBUG_TOOLBAR_ENABLED  # noqa: from local imports


 DEBUG_TOOLBAR_CONFIG = {
--- a/app/test_settings.py
+++ b/app/test_settings.py
@@ -0,0 +1,11 @@
+from app.settings import *  # noqa
+
+
+CACHES = {
+    "default": {
+        "BACKEND": "django.core.cache.backends.locmem.LocMemCache",
+        "LOCATION": "gia-test-cache",
+    }
+}
+
+CACHALOT_ENABLED = False
--- a/app/urls.py
+++ b/app/urls.py
@@ -13,15 +13,20 @@ Including another URLconf
    1. Import the include() function: from django.urls import include, path
    2. Add a URL to urlpatterns:  path('blog/', include('blog.urls'))
 """
+
 from django.conf import settings
 from django.conf.urls.static import static
 from django.contrib import admin
 from django.contrib.auth.views import LogoutView
 from django.urls import include, path
+from django.views.generic import RedirectView
 from two_factor.urls import urlpatterns as tf_urls
+from two_factor.views.profile import ProfileView

 from core.views import (
    ais,
+    automation,
+    availability,
    base,
    compose,
    groups,
@@ -33,33 +38,175 @@ from core.views import (
    osint,
    people,
    personas,
+    prosody,
    queues,
    sessions,
    signal,
    system,
+    tasks,
    whatsapp,
    workspace,
 )

 urlpatterns = [
-    path("__debug__/", include("debug_toolbar.urls")),
+    path(
+        "favicon.ico",
+        RedirectView.as_view(url=f"{settings.STATIC_URL}favicon.ico", permanent=False),
+    ),
    path("", base.Home.as_view(), name="home"),
    path("admin/", admin.site.urls),
+    path(
+        "account/two_factor/",
+        RedirectView.as_view(pattern_name="security_2fa", permanent=False),
+    ),
    # 2FA login urls
    path("", include(tf_urls)),
    path("accounts/signup/", base.Signup.as_view(), name="signup"),
    path("accounts/logout/", LogoutView.as_view(), name="logout"),
    # Notifications
+    path(
+        "notifications/page/update/",
+        RedirectView.as_view(pattern_name="notifications_settings", permanent=False),
+    ),
    path(
        "notifications/<str:type>/update/",
        notifications.NotificationsUpdate.as_view(),
        name="notifications_update",
    ),
+    path(
+        "settings/notifications/",
+        notifications.NotificationsUpdate.as_view(),
+        {"type": "page"},
+        name="notifications_settings",
+    ),
+    path(
+        "settings/security/",
+        system.SecurityPage.as_view(page_mode="encryption"),
+        name="security_settings",
+    ),
+    path(
+        "settings/security/encryption/",
+        system.SecurityPage.as_view(page_mode="encryption"),
+        name="encryption_settings",
+    ),
+    path(
+        "settings/security/permissions/",
+        system.SecurityPage.as_view(page_mode="permission"),
+        name="permission_settings",
+    ),
+    path(
+        "settings/security/permission/",
+        RedirectView.as_view(pattern_name="permission_settings", permanent=False),
+    ),
+    path(
+        "settings/security/2fa/",
+        ProfileView.as_view(),
+        name="security_2fa",
+    ),
+    path(
+        "settings/accessibility/",
+        system.AccessibilitySettings.as_view(),
+        name="accessibility_settings",
+    ),
+    path(
+        "settings/ai/",
+        system.AISettingsPage.as_view(),
+        name="ai_settings",
+    ),
+    path(
+        "settings/modules/",
+        system.ModulesSettingsPage.as_view(),
+        name="modules_settings",
+    ),
    path(
        "settings/system/",
        system.SystemSettings.as_view(),
        name="system_settings",
    ),
+    path(
+        "settings/system/capabilities/",
+        system.ServiceCapabilitySnapshotAPI.as_view(),
+        name="system_capabilities",
+    ),
+    path(
+        "settings/system/adapter-health/",
+        system.AdapterHealthSummaryAPI.as_view(),
+        name="system_adapter_health",
+    ),
+    path(
+        "settings/system/trace/",
+        system.TraceDiagnosticsAPI.as_view(),
+        name="system_trace_diagnostics",
+    ),
+    path(
+        "settings/system/projection-shadow/",
+        system.EventProjectionShadowAPI.as_view(),
+        name="system_projection_shadow",
+    ),
+    path(
+        "settings/system/event-ledger-smoke/",
+        system.EventLedgerSmokeAPI.as_view(),
+        name="system_event_ledger_smoke",
+    ),
+    path(
+        "settings/system/memory-search/status/",
+        system.MemorySearchStatusAPI.as_view(),
+        name="system_memory_search_status",
+    ),
+    path(
+        "settings/system/memory-search/query/",
+        system.MemorySearchQueryAPI.as_view(),
+        name="system_memory_search_query",
+    ),
+    path(
+        "internal/prosody/auth/",
+        prosody.ProsodyAuthBridge.as_view(),
+        name="prosody_auth_bridge",
+    ),
+    path(
+        "settings/command-routing/",
+        automation.CommandRoutingSettings.as_view(),
+        name="command_routing",
+    ),
+    path(
+        "settings/ai/traces/",
+        automation.AIExecutionLogSettings.as_view(),
+        name="ai_execution_log",
+    ),
+    path(
+        "settings/ai/traces/run/<int:run_id>/",
+        automation.AIExecutionRunDetailView.as_view(),
+        name="ai_execution_run_detail",
+    ),
+    path(
+        "settings/ai/traces/run/<int:run_id>/tab/<str:tab_slug>/",
+        automation.AIExecutionRunDetailTabView.as_view(),
+        name="ai_execution_run_detail_tab",
+    ),
+    path(
+        "settings/ai-execution/",
+        RedirectView.as_view(pattern_name="ai_execution_log", permanent=False),
+    ),
+    path(
+        "settings/translation/",
+        automation.TranslationSettings.as_view(),
+        name="translation_settings",
+    ),
+    path(
+        "settings/business-plans/",
+        automation.BusinessPlanInbox.as_view(),
+        name="business_plan_inbox",
+    ),
+    path(
+        "settings/business-plan/<str:doc_id>/",
+        automation.BusinessPlanEditor.as_view(),
+        name="business_plan_editor",
+    ),
+    path(
+        "settings/translation/preview/",
+        automation.TranslationPreview.as_view(),
+        name="translation_preview",
+    ),
    path(
        "services/signal/",
        signal.Signal.as_view(),
@@ -120,6 +267,11 @@ urlpatterns = [
        signal.SignalAccountAdd.as_view(),
        name="signal_account_add",
    ),
+    path(
+        "services/signal/<str:type>/unlink/<path:account>/",
+        signal.SignalAccountUnlink.as_view(),
+        name="signal_account_unlink",
+    ),
    path(
        "services/whatsapp/<str:type>/add/",
        whatsapp.WhatsAppAccountAdd.as_view(),
@@ -160,6 +312,11 @@ urlpatterns = [
        compose.ComposeSend.as_view(),
        name="compose_send",
    ),
+    path(
+        "compose/react/",
+        compose.ComposeReact.as_view(),
+        name="compose_react",
+    ),
    path(
        "compose/cancel-send/",
        compose.ComposeCancelSend.as_view(),
@@ -201,9 +358,14 @@ urlpatterns = [
        name="compose_thread",
    ),
    path(
-        "compose/history-sync/",
-        compose.ComposeHistorySync.as_view(),
-        name="compose_history_sync",
+        "compose/commands/bp/bind/",
+        compose.ComposeBindBP.as_view(),
+        name="compose_bind_bp",
+    ),
+    path(
+        "compose/commands/toggle/",
+        compose.ComposeToggleCommand.as_view(),
+        name="compose_toggle_command",
    ),
    path(
        "compose/media/blob/",
@@ -220,6 +382,76 @@ urlpatterns = [
        compose.ComposeContactMatch.as_view(),
        name="compose_contact_match",
    ),
+    path(
+        "compose/contacts/create/",
+        compose.ComposeContactCreate.as_view(),
+        name="compose_contact_create",
+    ),
+    path(
+        "compose/contacts/create-all/",
+        compose.ComposeContactCreateAll.as_view(),
+        name="compose_contact_create_all",
+    ),
+    path(
+        "compose/answer-suggestion/send/",
+        tasks.AnswerSuggestionSend.as_view(),
+        name="compose_answer_suggestion_send",
+    ),
+    path(
+        "tasks/",
+        tasks.TasksHub.as_view(),
+        name="tasks_hub",
+    ),
+    path(
+        "tasks/projects/<str:project_id>/",
+        tasks.TaskProjectDetail.as_view(),
+        name="tasks_project",
+    ),
+    path(
+        "tasks/epics/<str:epic_id>/",
+        tasks.TaskEpicDetail.as_view(),
+        name="tasks_epic",
+    ),
+    path(
+        "tasks/groups/<str:service>/<path:identifier>/",
+        tasks.TaskGroupDetail.as_view(),
+        name="tasks_group",
+    ),
+    path(
+        "tasks/task/<str:task_id>/",
+        tasks.TaskDetail.as_view(),
+        name="tasks_task",
+    ),
+    path(
+        "tasks/codex/submit/",
+        tasks.TaskCodexSubmit.as_view(),
+        name="tasks_codex_submit",
+    ),
+    path(
+        "settings/tasks/",
+        tasks.TaskSettings.as_view(),
+        name="tasks_settings",
+    ),
+    path(
+        "settings/codex/",
+        tasks.CodexSettingsPage.as_view(),
+        name="codex_settings",
+    ),
+    path(
+        "settings/codex/approval/",
+        tasks.CodexApprovalAction.as_view(),
+        name="codex_approval",
+    ),
+    path(
+        "settings/behavioral/",
+        availability.AvailabilitySettingsPage.as_view(),
+        name="behavioral_signals_settings",
+    ),
+    path(
+        "settings/availability/",
+        RedirectView.as_view(pattern_name="behavioral_signals_settings", permanent=False),
+        name="availability_settings",
+    ),
    # AIs
    path(
        "ai/workspace/",
@@ -341,6 +573,20 @@ urlpatterns = [
        workspace.AIWorkspaceUpdatePlanMeta.as_view(),
        name="ai_workspace_mitigation_meta_save",
    ),
+    path(
+        "settings/ai/models/",
+        ais.AIList.as_view(),
+        {"type": "page"},
+        name="ai_models",
+    ),
+    path(
+        "ai/models/",
+        RedirectView.as_view(pattern_name="ai_models", permanent=False),
+    ),
+    path(
+        "ai/page/",
+        RedirectView.as_view(pattern_name="ai_models", permanent=False),
+    ),
    path(
        "ai/<str:type>/",
        ais.AIList.as_view(),
@@ -556,3 +802,6 @@ urlpatterns = [
        name="queue_delete",
    ),
 ] + static(settings.STATIC_URL, document_root=settings.STATIC_ROOT)
+
+if getattr(settings, "DEBUG_TOOLBAR_ENABLED", False):
+    urlpatterns.insert(0, path("__debug__/", include("debug_toolbar.urls")))
--- a/artifacts/audits/1-initial.json
+++ b/artifacts/audits/1-initial.json
--- a/artifacts/audits/2-first-pass-fix.json
+++ b/artifacts/audits/2-first-pass-fix.json
@@ -0,0 +1,674 @@
+
+{
+  "score": 58,
+  "grade": "D",
+  "gradeLabel": "Significant security risks",
+  "totalFindings": 22,
+  "totalDepVulns": 0,
+  "categories": {
+    "secrets": {
+      "label": "Secrets",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "injection": {
+      "label": "Code Vulnerabilities",
+      "findingCount": 3,
+      "deduction": 9,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 3,
+        "low": 0
+      }
+    },
+    "deps": {
+      "label": "Dependencies",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "auth": {
+      "label": "Auth & Access Control",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "config": {
+      "label": "Configuration",
+      "findingCount": 4,
+      "deduction": 10,
+      "counts": {
+        "critical": 0,
+        "high": 4,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "supply-chain": {
+      "label": "Supply Chain",
+      "findingCount": 1,
+      "deduction": 3,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 1,
+        "low": 0
+      }
+    },
+    "api": {
+      "label": "API Security",
+      "findingCount": 3,
+      "deduction": 10,
+      "counts": {
+        "critical": 0,
+        "high": 3,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "llm": {
+      "label": "AI/LLM Security",
+      "findingCount": 11,
+      "deduction": 10,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 11,
+        "low": 0
+      }
+    }
+  },
+  "findings": [
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 25,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    },
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 28,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    },
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 30,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    },
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 31,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/clients/whatsapp.py",
+      "line": 3398,
+      "severity": "high",
+      "category": "api",
+      "rule": "API_UPLOAD_NO_TYPE_CHECK",
+      "title": "API: File Upload Without Type Validation",
+      "description": "File upload using original filename without type validation.",
+      "fix": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "cwe": "CWE-434",
+      "owasp": "A04:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/clients/xmpp.py",
+      "line": 57,
+      "severity": "high",
+      "category": "api",
+      "rule": "API_UPLOAD_NO_TYPE_CHECK",
+      "title": "API: File Upload Without Type Validation",
+      "description": "File upload using original filename without type validation.",
+      "fix": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "cwe": "CWE-434",
+      "owasp": "A04:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/security/attachments.py",
+      "line": 83,
+      "severity": "high",
+      "category": "api",
+      "rule": "API_UPLOAD_NO_TYPE_CHECK",
+      "title": "API: File Upload Without Type Validation",
+      "description": "File upload using original filename without type validation.",
+      "fix": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "cwe": "CWE-434",
+      "owasp": "A04:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/tests/test_attachment_security.py",
+      "line": 29,
+      "severity": "medium",
+      "category": "ssrf",
+      "rule": "SSRF_INTERNAL_IP",
+      "title": "SSRF: Internal IP Pattern",
+      "description": "Internal IP address in code. Verify it is not reachable via user-controlled URLs.",
+      "fix": "Block private IP ranges in URL validation for user-supplied URLs",
+      "cwe": "CWE-918",
+      "owasp": "A10:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/tests/test_attachment_security.py",
+      "line": 34,
+      "severity": "medium",
+      "category": "ssrf",
+      "rule": "SSRF_INTERNAL_IP",
+      "title": "SSRF: Internal IP Pattern",
+      "description": "Internal IP address in code. Verify it is not reachable via user-controlled URLs.",
+      "fix": "Block private IP ranges in URL validation for user-supplied URLs",
+      "cwe": "CWE-918",
+      "owasp": "A10:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/tests/test_attachment_security.py",
+      "line": 35,
+      "severity": "medium",
+      "category": "ssrf",
+      "rule": "SSRF_INTERNAL_IP",
+      "title": "SSRF: Internal IP Pattern",
+      "description": "Internal IP address in code. Verify it is not reachable via user-controlled URLs.",
+      "fix": "Block private IP ranges in URL validation for user-supplied URLs",
+      "cwe": "CWE-918",
+      "owasp": "A10:2021"
+    },
+    {
+      "file": "/code/xf/GIA/requirements.txt",
+      "line": 23,
+      "severity": "medium",
+      "category": "supply-chain",
+      "rule": "UNPINNED_PYTHON_DEP",
+      "title": "Unpinned Python Dependency: ./vendor/django-crud-mixins",
+      "description": "Python dependency without version pin. Pin to a specific version for reproducible builds.",
+      "fix": "Pin version: ./vendor/django-crud-mixins==x.y.z",
+      "cwe": null,
+      "owasp": null
+    },
+    {
+      "file": "/code/xf/GIA/core/clients/signalapi.py",
+      "line": 411,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_NO_OUTPUT_FILTER",
+      "title": "LLM Output Without Filtering",
+      "description": "LLM output used directly without filtering. May contain sensitive info or hallucinations.",
+      "fix": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "cwe": "CWE-200",
+      "owasp": "LLM02"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 739,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 744,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 758,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 850,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 1377,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 1382,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 1396,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/signal.py",
+      "line": 189,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_NO_OUTPUT_FILTER",
+      "title": "LLM Output Without Filtering",
+      "description": "LLM output used directly without filtering. May contain sensitive info or hallucinations.",
+      "fix": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "cwe": "CWE-200",
+      "owasp": "LLM02"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/signal.py",
+      "line": 197,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_NO_OUTPUT_FILTER",
+      "title": "LLM Output Without Filtering",
+      "description": "LLM output used directly without filtering. May contain sensitive info or hallucinations.",
+      "fix": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "cwe": "CWE-200",
+      "owasp": "LLM02"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/signal.py",
+      "line": 206,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_NO_OUTPUT_FILTER",
+      "title": "LLM Output Without Filtering",
+      "description": "LLM output used directly without filtering. May contain sensitive info or hallucinations.",
+      "fix": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "cwe": "CWE-200",
+      "owasp": "LLM02"
+    }
+  ],
+  "depVulns": [],
+  "remediationPlan": [
+    {
+      "priority": 1,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:25",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    },
+    {
+      "priority": 2,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:28",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    },
+    {
+      "priority": 3,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:30",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    },
+    {
+      "priority": 4,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:31",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    },
+    {
+      "priority": 5,
+      "severity": "high",
+      "category": "api",
+      "categoryLabel": "API SECURITY",
+      "title": "API: File Upload Without Type Validation",
+      "file": "core/clients/whatsapp.py:3398",
+      "action": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "effort": "medium"
+    },
+    {
+      "priority": 6,
+      "severity": "high",
+      "category": "api",
+      "categoryLabel": "API SECURITY",
+      "title": "API: File Upload Without Type Validation",
+      "file": "core/clients/xmpp.py:57",
+      "action": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "effort": "medium"
+    },
+    {
+      "priority": 7,
+      "severity": "high",
+      "category": "api",
+      "categoryLabel": "API SECURITY",
+      "title": "API: File Upload Without Type Validation",
+      "file": "core/security/attachments.py:83",
+      "action": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "effort": "medium"
+    },
+    {
+      "priority": 8,
+      "severity": "medium",
+      "category": "ssrf",
+      "categoryLabel": "SSRF",
+      "title": "SSRF: Internal IP Pattern",
+      "file": "core/tests/test_attachment_security.py:29",
+      "action": "Block private IP ranges in URL validation for user-supplied URLs",
+      "effort": "medium"
+    },
+    {
+      "priority": 9,
+      "severity": "medium",
+      "category": "ssrf",
+      "categoryLabel": "SSRF",
+      "title": "SSRF: Internal IP Pattern",
+      "file": "core/tests/test_attachment_security.py:34",
+      "action": "Block private IP ranges in URL validation for user-supplied URLs",
+      "effort": "medium"
+    },
+    {
+      "priority": 10,
+      "severity": "medium",
+      "category": "ssrf",
+      "categoryLabel": "SSRF",
+      "title": "SSRF: Internal IP Pattern",
+      "file": "core/tests/test_attachment_security.py:35",
+      "action": "Block private IP ranges in URL validation for user-supplied URLs",
+      "effort": "medium"
+    },
+    {
+      "priority": 11,
+      "severity": "medium",
+      "category": "supply-chain",
+      "categoryLabel": "SUPPLY CHAIN",
+      "title": "Unpinned Python Dependency: ./vendor/django-crud-mixins",
+      "file": "requirements.txt:23",
+      "action": "Pin version: ./vendor/django-crud-mixins==x.y.z",
+      "effort": "medium"
+    },
+    {
+      "priority": 12,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "LLM Output Without Filtering",
+      "file": "core/clients/signalapi.py:411",
+      "action": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "effort": "high"
+    },
+    {
+      "priority": 13,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:739",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 14,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:744",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 15,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:758",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 16,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:850",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 17,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:1377",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 18,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:1382",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 19,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:1396",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 20,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "LLM Output Without Filtering",
+      "file": "core/views/signal.py:189",
+      "action": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "effort": "high"
+    },
+    {
+      "priority": 21,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "LLM Output Without Filtering",
+      "file": "core/views/signal.py:197",
+      "action": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "effort": "high"
+    },
+    {
+      "priority": 22,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "LLM Output Without Filtering",
+      "file": "core/views/signal.py:206",
+      "action": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "effort": "high"
+    }
+  ],
+  "recon": {
+    "frameworks": [
+      "django"
+    ],
+    "languages": [
+      "python"
+    ],
+    "apiRoutes": [
+      "app/urls.py",
+      "core/management/commands/backfill_xmpp_attachment_urls.py"
+    ],
+    "authPatterns": [],
+    "databases": [],
+    "cloudProviders": [],
+    "frontendExposure": [],
+    "packageManagers": [
+      "pip"
+    ],
+    "cicd": [],
+    "hasDockerfile": true,
+    "hasTerraform": false,
+    "hasKubernetes": false,
+    "envFiles": [],
+    "configFiles": []
+  },
+  "agents": [
+    {
+      "agent": "InjectionTester",
+      "category": "injection",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "AuthBypassAgent",
+      "category": "auth",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SSRFProber",
+      "category": "ssrf",
+      "findingCount": 3,
+      "success": true
+    },
+    {
+      "agent": "SupplyChainAudit",
+      "category": "supply-chain",
+      "findingCount": 1,
+      "success": true
+    },
+    {
+      "agent": "ConfigAuditor",
+      "category": "config",
+      "findingCount": 4,
+      "success": true
+    },
+    {
+      "agent": "LLMRedTeam",
+      "category": "llm",
+      "findingCount": 11,
+      "success": true
+    },
+    {
+      "agent": "MobileScanner",
+      "category": "mobile",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "GitHistoryScanner",
+      "category": "history",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "CICDScanner",
+      "category": "cicd",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "APIFuzzer",
+      "category": "api",
+      "findingCount": 3,
+      "success": true
+    }
+  ]
+}
--- a/artifacts/audits/3-second-pass-fix.json
+++ b/artifacts/audits/3-second-pass-fix.json
@@ -0,0 +1,410 @@
+
+{
+  "score": 74,
+  "grade": "C",
+  "gradeLabel": "Fix before shipping",
+  "totalFindings": 10,
+  "totalDepVulns": 0,
+  "categories": {
+    "secrets": {
+      "label": "Secrets",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "injection": {
+      "label": "Code Vulnerabilities",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "deps": {
+      "label": "Dependencies",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "auth": {
+      "label": "Auth & Access Control",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "config": {
+      "label": "Configuration",
+      "findingCount": 1,
+      "deduction": 8,
+      "counts": {
+        "critical": 0,
+        "high": 1,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "supply-chain": {
+      "label": "Supply Chain",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "api": {
+      "label": "API Security",
+      "findingCount": 1,
+      "deduction": 8,
+      "counts": {
+        "critical": 0,
+        "high": 1,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "llm": {
+      "label": "AI/LLM Security",
+      "findingCount": 8,
+      "deduction": 10,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 8,
+        "low": 0
+      }
+    }
+  },
+  "findings": [
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 26,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/security/attachments.py",
+      "line": 113,
+      "severity": "high",
+      "category": "api",
+      "rule": "API_UPLOAD_NO_TYPE_CHECK",
+      "title": "API: File Upload Without Type Validation",
+      "description": "File upload using original filename without type validation.",
+      "fix": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "cwe": "CWE-434",
+      "owasp": "A04:2021"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 775,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 781,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 795,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 1418,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 1424,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/osint.py",
+      "line": 1438,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_RAG_NO_VALIDATION",
+      "title": "RAG Pipeline Without Input Validation",
+      "description": "User input passed directly to vector search/embedding without validation.",
+      "fix": "Validate and sanitize input before embedding. Limit query length.",
+      "cwe": "CWE-20",
+      "owasp": "LLM08"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/signal.py",
+      "line": 202,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_NO_OUTPUT_FILTER",
+      "title": "LLM Output Without Filtering",
+      "description": "LLM output used directly without filtering. May contain sensitive info or hallucinations.",
+      "fix": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "cwe": "CWE-200",
+      "owasp": "LLM02"
+    },
+    {
+      "file": "/code/xf/GIA/core/views/signal.py",
+      "line": 211,
+      "severity": "medium",
+      "category": "llm",
+      "rule": "LLM_NO_OUTPUT_FILTER",
+      "title": "LLM Output Without Filtering",
+      "description": "LLM output used directly without filtering. May contain sensitive info or hallucinations.",
+      "fix": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "cwe": "CWE-200",
+      "owasp": "LLM02"
+    }
+  ],
+  "depVulns": [],
+  "remediationPlan": [
+    {
+      "priority": 1,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:26",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    },
+    {
+      "priority": 2,
+      "severity": "high",
+      "category": "api",
+      "categoryLabel": "API SECURITY",
+      "title": "API: File Upload Without Type Validation",
+      "file": "core/security/attachments.py:113",
+      "action": "Validate file extension and MIME type. Generate random filenames for storage.",
+      "effort": "medium"
+    },
+    {
+      "priority": 3,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:775",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 4,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:781",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 5,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:795",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 6,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:1418",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 7,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:1424",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 8,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "RAG Pipeline Without Input Validation",
+      "file": "core/views/osint.py:1438",
+      "action": "Validate and sanitize input before embedding. Limit query length.",
+      "effort": "high"
+    },
+    {
+      "priority": 9,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "LLM Output Without Filtering",
+      "file": "core/views/signal.py:202",
+      "action": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "effort": "high"
+    },
+    {
+      "priority": 10,
+      "severity": "medium",
+      "category": "llm",
+      "categoryLabel": "AI/LLM SECURITY",
+      "title": "LLM Output Without Filtering",
+      "file": "core/views/signal.py:211",
+      "action": "Filter LLM output before displaying: remove PII, validate against expected format",
+      "effort": "high"
+    }
+  ],
+  "recon": {
+    "frameworks": [
+      "django"
+    ],
+    "languages": [
+      "python"
+    ],
+    "apiRoutes": [
+      "app/urls.py",
+      "core/management/commands/backfill_xmpp_attachment_urls.py"
+    ],
+    "authPatterns": [],
+    "databases": [],
+    "cloudProviders": [],
+    "frontendExposure": [],
+    "packageManagers": [
+      "pip"
+    ],
+    "cicd": [],
+    "hasDockerfile": true,
+    "hasTerraform": false,
+    "hasKubernetes": false,
+    "envFiles": [],
+    "configFiles": []
+  },
+  "agents": [
+    {
+      "agent": "InjectionTester",
+      "category": "injection",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "AuthBypassAgent",
+      "category": "auth",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SSRFProber",
+      "category": "ssrf",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SupplyChainAudit",
+      "category": "supply-chain",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "ConfigAuditor",
+      "category": "config",
+      "findingCount": 1,
+      "success": true
+    },
+    {
+      "agent": "LLMRedTeam",
+      "category": "llm",
+      "findingCount": 8,
+      "success": true
+    },
+    {
+      "agent": "MobileScanner",
+      "category": "mobile",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "GitHistoryScanner",
+      "category": "history",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "CICDScanner",
+      "category": "cicd",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "APIFuzzer",
+      "category": "api",
+      "findingCount": 1,
+      "success": true
+    }
+  ]
+}
--- a/artifacts/audits/4-third-pass-fix.json
+++ b/artifacts/audits/4-third-pass-fix.json
@@ -0,0 +1,212 @@
+
+{
+  "score": 92,
+  "grade": "A",
+  "gradeLabel": "Ship it!",
+  "totalFindings": 1,
+  "totalDepVulns": 0,
+  "categories": {
+    "secrets": {
+      "label": "Secrets",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "injection": {
+      "label": "Code Vulnerabilities",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "deps": {
+      "label": "Dependencies",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "auth": {
+      "label": "Auth & Access Control",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "config": {
+      "label": "Configuration",
+      "findingCount": 1,
+      "deduction": 8,
+      "counts": {
+        "critical": 0,
+        "high": 1,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "supply-chain": {
+      "label": "Supply Chain",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "api": {
+      "label": "API Security",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "llm": {
+      "label": "AI/LLM Security",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    }
+  },
+  "findings": [
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 26,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    }
+  ],
+  "depVulns": [],
+  "remediationPlan": [
+    {
+      "priority": 1,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:26",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    }
+  ],
+  "recon": {
+    "frameworks": [
+      "django"
+    ],
+    "languages": [
+      "python"
+    ],
+    "apiRoutes": [
+      "app/urls.py",
+      "core/management/commands/backfill_xmpp_attachment_urls.py"
+    ],
+    "authPatterns": [],
+    "databases": [],
+    "cloudProviders": [],
+    "frontendExposure": [],
+    "packageManagers": [
+      "pip"
+    ],
+    "cicd": [],
+    "hasDockerfile": true,
+    "hasTerraform": false,
+    "hasKubernetes": false,
+    "envFiles": [],
+    "configFiles": []
+  },
+  "agents": [
+    {
+      "agent": "InjectionTester",
+      "category": "injection",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "AuthBypassAgent",
+      "category": "auth",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SSRFProber",
+      "category": "ssrf",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SupplyChainAudit",
+      "category": "supply-chain",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "ConfigAuditor",
+      "category": "config",
+      "findingCount": 1,
+      "success": true
+    },
+    {
+      "agent": "LLMRedTeam",
+      "category": "llm",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "MobileScanner",
+      "category": "mobile",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "GitHistoryScanner",
+      "category": "history",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "CICDScanner",
+      "category": "cicd",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "APIFuzzer",
+      "category": "api",
+      "findingCount": 0,
+      "success": true
+    }
+  ]
+}
--- a/artifacts/audits/5-final-pass-fix.json
+++ b/artifacts/audits/5-final-pass-fix.json
@@ -0,0 +1,234 @@
+
+{
+  "score": 90,
+  "grade": "A",
+  "gradeLabel": "Ship it!",
+  "totalFindings": 2,
+  "totalDepVulns": 0,
+  "categories": {
+    "secrets": {
+      "label": "Secrets",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "injection": {
+      "label": "Code Vulnerabilities",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "deps": {
+      "label": "Dependencies",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "auth": {
+      "label": "Auth & Access Control",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "config": {
+      "label": "Configuration",
+      "findingCount": 2,
+      "deduction": 10,
+      "counts": {
+        "critical": 0,
+        "high": 2,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "supply-chain": {
+      "label": "Supply Chain",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "api": {
+      "label": "API Security",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    },
+    "llm": {
+      "label": "AI/LLM Security",
+      "findingCount": 0,
+      "deduction": 0,
+      "counts": {
+        "critical": 0,
+        "high": 0,
+        "medium": 0,
+        "low": 0
+      }
+    }
+  },
+  "findings": [
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 26,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_RUN_AS_ROOT",
+      "title": "Docker: Running as Root",
+      "description": "No USER instruction found. Container runs as root by default.",
+      "fix": "Add USER nonroot before CMD/ENTRYPOINT",
+      "cwe": "CWE-250",
+      "owasp": "A05:2021"
+    },
+    {
+      "file": "/code/xf/GIA/Dockerfile",
+      "line": 1,
+      "severity": "high",
+      "category": "config",
+      "rule": "DOCKER_NO_USER",
+      "title": "Dockerfile: No Non-Root USER",
+      "description": "No USER instruction found. Container runs as root, enabling escape attacks.",
+      "fix": "Add before CMD: RUN addgroup -S app && adduser -S app -G app\nUSER app",
+      "cwe": null,
+      "owasp": null
+    }
+  ],
+  "depVulns": [],
+  "remediationPlan": [
+    {
+      "priority": 1,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Docker: Running as Root",
+      "file": "Dockerfile:26",
+      "action": "Add USER nonroot before CMD/ENTRYPOINT",
+      "effort": "low"
+    },
+    {
+      "priority": 2,
+      "severity": "high",
+      "category": "config",
+      "categoryLabel": "CONFIGURATION",
+      "title": "Dockerfile: No Non-Root USER",
+      "file": "Dockerfile:1",
+      "action": "Add before CMD: RUN addgroup -S app && adduser -S app -G app\nUSER app",
+      "effort": "low"
+    }
+  ],
+  "recon": {
+    "frameworks": [
+      "django"
+    ],
+    "languages": [
+      "python"
+    ],
+    "apiRoutes": [
+      "app/urls.py",
+      "core/management/commands/backfill_xmpp_attachment_urls.py"
+    ],
+    "authPatterns": [],
+    "databases": [],
+    "cloudProviders": [],
+    "frontendExposure": [],
+    "packageManagers": [
+      "pip"
+    ],
+    "cicd": [],
+    "hasDockerfile": true,
+    "hasTerraform": false,
+    "hasKubernetes": false,
+    "envFiles": [],
+    "configFiles": []
+  },
+  "agents": [
+    {
+      "agent": "InjectionTester",
+      "category": "injection",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "AuthBypassAgent",
+      "category": "auth",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SSRFProber",
+      "category": "ssrf",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "SupplyChainAudit",
+      "category": "supply-chain",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "ConfigAuditor",
+      "category": "config",
+      "findingCount": 2,
+      "success": true
+    },
+    {
+      "agent": "LLMRedTeam",
+      "category": "llm",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "MobileScanner",
+      "category": "mobile",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "GitHistoryScanner",
+      "category": "history",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "CICDScanner",
+      "category": "cicd",
+      "findingCount": 0,
+      "success": true
+    },
+    {
+      "agent": "APIFuzzer",
+      "category": "api",
+      "findingCount": 0,
+      "success": true
+    }
+  ]
+}
--- a/artifacts/frontend_asset_playbook.md
+++ b/artifacts/frontend_asset_playbook.md
@@ -0,0 +1,87 @@
+# Frontend Asset Vendoring Playbook
+
+This is the repeatable process used in GIA to self-host third-party frontend assets, pin versions, record provenance, and keep template SRI hashes maintainable.
+
+## Goals
+
+- Avoid runtime CDN dependencies for core UI assets.
+- Pin exact upstream versions and package integrity in source control.
+- Record provenance, license, and SRI for each third-party asset.
+- Keep a simple manifest-driven workflow that can be copied into other projects.
+
+## Files
+
+- `tools/frontend_assets/package.json`
+  - Exact npm dependency pins for vendored frontend libraries.
+- `tools/frontend_assets/package-lock.json`
+  - npm lockfile integrity and tarball resolution source of truth.
+- `tools/frontend_assets/asset-manifest.json`
+  - Maps upstream package files and URL bundles into local `core/static/` targets.
+- `scripts/vendor_frontend_assets.py`
+  - Copies npm assets, downloads non-npm bundles, computes local SHA-256 and SRI SHA-512, and emits inventory reports.
+- `artifacts/frontend_libraries.md`
+  - Human-readable inventory with versions, licenses, official URLs, tarballs, upstream integrity, and local SHA/SRI.
+- `artifacts/frontend_libraries.json`
+  - Machine-readable inventory for template updates and audits.
+
+## Standard Workflow
+
+1. Add or update exact versions in `tools/frontend_assets/package.json`.
+2. Confirm each package is the official distribution, not a typosquat or community fork.
+3. Update `tools/frontend_assets/asset-manifest.json` with the exact files to vendor.
+4. Run `npm install --ignore-scripts` in `tools/frontend_assets/`.
+5. Run `./genv/bin/python scripts/vendor_frontend_assets.py`.
+6. Copy the emitted `sri_sha512` values into template tags in `core/templates/base.html`.
+7. Restart the stack so `collectstatic` republishes the new files.
+8. Verify in the browser that assets load from local static URLs with no CDN fallbacks.
+
+## Provenance Rules
+
+- Prefer official project docs or official extension listings before trusting npm search results.
+- For Bulma extensions, treat `https://bulma.io/extensions/` as the discovery source and verify the linked package/repo matches the npm package name.
+- Confirm npm metadata matches the expected project:
+  - homepage
+  - repository
+  - maintainers
+  - license
+  - publish recency
+- Keep exact versions in `package.json`; do not use ranges.
+- Use `package-lock.json` as the immutable record of tarball URLs and upstream integrity.
+- If a library is not safely available from npm, vendor it from the official upstream URL as a `url_bundle`.
+- Do not silently substitute a “free” or differently-scoped package if the templates rely on another asset family.
+
+## GIA Decisions
+
+- Font Awesome:
+  - GIA templates use icon classes that are not safely replaceable with `@fortawesome/fontawesome-free`.
+  - The existing Font Awesome Pro `site-assets` v6.1.1 bundle is self-hosted under `core/static/vendor/fontawesome/` instead.
+- jQuery:
+  - Latest npm is 4.x, but GIA stays on `3.7.1` to avoid breaking older plugins.
+- Bulma extensions:
+  - `bulma-calendar`, `bulma-tagsinput`, `bulma-switch`, `bulma-slider`, and `bulma-tooltip` were matched against Bulma's official extensions page before pinning.
+  - `bulma-calendar` and `bulma-tooltip` are deprecated on npm, but Bulma still points to the Wikiki ecosystem for these extensions, so they were kept and documented instead of replaced ad hoc.
+
+## Theme Strategy
+
+- `core/static/css/gia-theme.css` is now the project override layer instead of an inline `<style>` block in `base.html`.
+- This is the low-friction step that works with the current stack immediately.
+- If we want a true Bulma theme build later, add `sass` to `tools/frontend_assets`, create a `gia-bulma.scss` entrypoint that sets Bulma theme variables before importing Bulma, and vendor the compiled output through the same manifest process.
+
+## Commands
+
+```bash
+cd /code/xf/GIA/tools/frontend_assets
+npm install --ignore-scripts
+
+cd /code/xf/GIA
+./genv/bin/python scripts/vendor_frontend_assets.py
+bash ./scripts/quadlet/manage.sh restart
+./genv/bin/python manage.py check
+```
+
+## Verification Checklist
+
+- `core/templates/base.html` only references local static URLs for vendored libraries.
+- Each included stylesheet/script in `base.html` has an `integrity="sha512-..."` attribute.
+- Browser network panel shows local asset hits and no unexpected CDN requests.
+- `artifacts/frontend_libraries.md` reflects the exact deployed asset set.
--- a/artifacts/frontend_libraries.json
+++ b/artifacts/frontend_libraries.json
@@ -0,0 +1,469 @@
+{
+  "entries": [
+    {
+      "id": "bulma_css",
+      "kind": "npm_file",
+      "package": "bulma",
+      "version": "1.0.4",
+      "license": "MIT",
+      "homepage": "https://bulma.io",
+      "official_url": "https://bulma.io",
+      "purpose": "Bulma core stylesheet",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma/-/bulma-1.0.4.tgz",
+      "dist_integrity": "sha512-Ffb6YGXDiZYX3cqvSbHWqQ8+LkX6tVoTcZuVB3lm93sbAVXlO0D6QlOTMnV6g18gILpAXqkG2z9hf9z4hCjz2g==",
+      "source_path": "css/bulma.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/bulma.min.css",
+          "sha256": "67fa26df1ca9e95d8f2adc7c04fa1b15fa3d24257470ebc10cc68b9aab914bee",
+          "sri_sha512": "sha512-yh2RE0wZCVZeysGiqTwDTO/dKelCbS9bP2L94UvOFtl/FKXcNAje3Y2oBg/ZMZ3LS1sicYk4dYVGtDex75fvvA=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_tooltip_css",
+      "kind": "npm_file",
+      "package": "bulma-tooltip",
+      "version": "3.0.2",
+      "license": "MIT",
+      "homepage": "https://github.com/Wikiki/bulma-tooltip#readme",
+      "official_url": "https://wikiki.github.io/elements/tooltip",
+      "purpose": "Official Bulma tooltip extension from Bulma's extensions page",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-tooltip/-/bulma-tooltip-3.0.2.tgz",
+      "dist_integrity": "sha512-CsT3APjhlZScskFg38n8HYL8oYNUHQtcu4sz6ERarxkUpBRbk9v0h/5KAvXeKapVSn2dp9l7bOGit5SECP8EWQ==",
+      "source_path": "dist/css/bulma-tooltip.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/bulma-tooltip.min.css",
+          "sha256": "5c79d12a40b3532aaec159faa0b85fd3d500e192467761b71e0bda0fd04f3076",
+          "sri_sha512": "sha512-SNDNIUvSYhnqDV9FFXaH/e0xZ6NzkG4Qm5dafLLf0PCMkzICKaOmMTgI3y2t2jZK+hAtP6A7UBcFqjWMhsujIg=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_slider_css",
+      "kind": "npm_file",
+      "package": "bulma-slider",
+      "version": "2.0.5",
+      "license": "MIT",
+      "homepage": "https://github.com/Wikiki/bulma-slider#readme",
+      "official_url": "https://wikiki.github.io/form/slider",
+      "purpose": "Official Bulma slider extension from Bulma's extensions page",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-slider/-/bulma-slider-2.0.5.tgz",
+      "dist_integrity": "sha512-6woD/1E7q1o5bfEaQjNqpWZaCItC1oHe9bN15WYB2ELqz2gDaJYZkf+rlozGpAYOXQGDQGCCv3y+QuKjx6sQuw==",
+      "source_path": "dist/css/bulma-slider.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/bulma-slider.min.css",
+          "sha256": "f9d952627d388b8ba267e1388d6923274cf9e62e758d459c5a045f3933e9dc8a",
+          "sri_sha512": "sha512-9o5SkCRCA9thttRH3Gb5QXLxKdRiuRLdO6ToEPwRHGLXjrhTZwFj0rEHjrCcJvDN9/aNaWMpGOIEA2vZsHmEqw=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_slider_js",
+      "kind": "npm_file",
+      "package": "bulma-slider",
+      "version": "2.0.5",
+      "license": "MIT",
+      "homepage": "https://github.com/Wikiki/bulma-slider#readme",
+      "official_url": "https://wikiki.github.io/form/slider",
+      "purpose": "Official Bulma slider extension runtime",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-slider/-/bulma-slider-2.0.5.tgz",
+      "dist_integrity": "sha512-6woD/1E7q1o5bfEaQjNqpWZaCItC1oHe9bN15WYB2ELqz2gDaJYZkf+rlozGpAYOXQGDQGCCv3y+QuKjx6sQuw==",
+      "source_path": "dist/js/bulma-slider.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/bulma-slider.min.js",
+          "sha256": "db68ebe154a25597913c5635f31500fe7a32e5a205fb9a98c9642d0c2de47d9e",
+          "sri_sha512": "sha512-WLKXHCsMXTSIPsmQShJRE6K4IzwvNkhwxr/Oo8N3z+kzjhGleHibspmWLTawNMdl2z9E23XK20+yvUTDZ+zeNQ=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_calendar_css",
+      "kind": "npm_file",
+      "package": "bulma-calendar",
+      "version": "7.1.1",
+      "license": "MIT",
+      "homepage": "https://doc.mh-s.de/bulma-calendar",
+      "official_url": "https://wikiki.github.io/components/calendar",
+      "purpose": "Official Bulma calendar extension from Bulma's extensions page",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-calendar/-/bulma-calendar-7.1.1.tgz",
+      "dist_integrity": "sha512-E08i25KOfqMKBndgDF3y3eoQ0dUzVkgV9R53EDRM65GQUQKLzt8gcXVJYs3mYnpq6L3DiLuUt47Fl09tSv9OpA==",
+      "source_path": "src/demo/assets/css/bulma-calendar.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/bulma-calendar.min.css",
+          "sha256": "d18b488ca52584bcd6ea3fb84bf06380e47a3cd18660a235617da017d13ab269",
+          "sri_sha512": "sha512-IOnJQkgQpezPDPTJcRiWD7YVI3sF2RYzYDl4isbDT2geSaEHRQ615UN/8GhJbSkvqkKRZu8SBCQ7XwKMqsqLFQ=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_calendar_js",
+      "kind": "npm_file",
+      "package": "bulma-calendar",
+      "version": "7.1.1",
+      "license": "MIT",
+      "homepage": "https://doc.mh-s.de/bulma-calendar",
+      "official_url": "https://wikiki.github.io/components/calendar",
+      "purpose": "Official Bulma calendar extension runtime",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-calendar/-/bulma-calendar-7.1.1.tgz",
+      "dist_integrity": "sha512-E08i25KOfqMKBndgDF3y3eoQ0dUzVkgV9R53EDRM65GQUQKLzt8gcXVJYs3mYnpq6L3DiLuUt47Fl09tSv9OpA==",
+      "source_path": "src/demo/assets/js/bulma-calendar.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/bulma-calendar.min.js",
+          "sha256": "58160c87c4d17f9d98ec366fe019492acde50efbc0297af7045547952b306680",
+          "sri_sha512": "sha512-kkEtEtypXzruevjkoxhyEkqkZBtlhK7s8zt7IV2yPabgBwy5xbKL9uWeCS37ldS9AaNTSnveWTu4ivUvGMJUWA=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_tagsinput_css",
+      "kind": "npm_file",
+      "package": "bulma-tagsinput",
+      "version": "2.0.0",
+      "license": "MIT",
+      "homepage": "https://github.com/Wikiki/bulma-tagsinput#readme",
+      "official_url": "https://wikiki.github.io/form/tagsinput",
+      "purpose": "Official Bulma tagsinput extension from Bulma's extensions page",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-tagsinput/-/bulma-tagsinput-2.0.0.tgz",
+      "dist_integrity": "sha512-BFvd0oaxgeWHOEh3d4cgETy5vpSSjRRBA9w+8TWEuhjFQg38Rb+3vjDCavL+udpdjf+dRV0SK5T4kYCXTOrz5A==",
+      "source_path": "dist/css/bulma-tagsinput.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/bulma-tagsinput.min.css",
+          "sha256": "8d1de24619c05ddf9045638b52059ab492d4887ce74119eed545d66af859da89",
+          "sri_sha512": "sha512-NWTkcDRubZ3pyXbZZLQBILuVsRFs8c6QGgnfe4dm5/d6yp50U+xdoCDLIcSo51fFy/GXH0O2Oed1Z1sF1faxDA=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_tagsinput_js",
+      "kind": "npm_file",
+      "package": "bulma-tagsinput",
+      "version": "2.0.0",
+      "license": "MIT",
+      "homepage": "https://github.com/Wikiki/bulma-tagsinput#readme",
+      "official_url": "https://wikiki.github.io/form/tagsinput",
+      "purpose": "Official Bulma tagsinput extension runtime",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-tagsinput/-/bulma-tagsinput-2.0.0.tgz",
+      "dist_integrity": "sha512-BFvd0oaxgeWHOEh3d4cgETy5vpSSjRRBA9w+8TWEuhjFQg38Rb+3vjDCavL+udpdjf+dRV0SK5T4kYCXTOrz5A==",
+      "source_path": "dist/js/bulma-tagsinput.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/bulma-tagsinput.min.js",
+          "sha256": "b355aa94ec519e374d7edf569e3dbde8bbe30ff3a193cb96f2930ee7815939d6",
+          "sri_sha512": "sha512-Je6J++MjmmpxF30JCmRwM2KiK3uWQBQtqiNCjwzEMJKExLaa0BqerlYNa/fJAl5Rra4hMgRZF2fzg+V2vjE4Kw=="
+        }
+      ]
+    },
+    {
+      "id": "bulma_switch_css",
+      "kind": "npm_file",
+      "package": "bulma-switch",
+      "version": "2.0.4",
+      "license": "MIT",
+      "homepage": "https://github.com/Wikiki/bulma-switch#readme",
+      "official_url": "https://wikiki.github.io/form/switch",
+      "purpose": "Official Bulma switch extension from Bulma's extensions page",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/bulma-switch/-/bulma-switch-2.0.4.tgz",
+      "dist_integrity": "sha512-kMu4H0Pr0VjvfsnT6viRDCgptUq0Rvy7y7PX6q+IHg1xUynsjszPjhAdal5ysAlCG5HNO+5YXxeiu92qYGQolw==",
+      "source_path": "dist/css/bulma-switch.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/bulma-switch.min.css",
+          "sha256": "f0460ddebdd95425a50590908503a170f5ff08b28bd53573c71791fc7cd1e6f5",
+          "sri_sha512": "sha512-zjrHYubQoNgDVqVKTyGjKcvIeQlduZTvXCvcBwQ0iqJYKLKiz9cuFAN7e98zfKqCTpI/EgFRBRcTwJw20yAFuw=="
+        }
+      ]
+    },
+    {
+      "id": "gridstack_css",
+      "kind": "npm_file",
+      "package": "gridstack",
+      "version": "12.4.2",
+      "license": "MIT",
+      "homepage": "http://gridstackjs.com/",
+      "official_url": "https://gridstackjs.com/",
+      "purpose": "GridStack stylesheet",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/gridstack/-/gridstack-12.4.2.tgz",
+      "dist_integrity": "sha512-aXbJrQpi3LwpYXYOr4UriPM5uc/dPcjK01SdOE5PDpx2vi8tnLhU7yBg/1i4T59UhNkG/RBfabdFUObuN+gMnw==",
+      "source_path": "dist/gridstack.min.css",
+      "targets": [
+        {
+          "path": "core/static/css/gridstack.min.css",
+          "sha256": "55e9d4ea6d8c6f8f1ea8a449b8af18d8571487c0afc6b433cccf877047cb8457",
+          "sri_sha512": "sha512-ttQfsDTO64bamkJHeLDf0kzMP1NKfkootudPWS2V8Pwy+9z1wexSYjIT6/HXGg/bmtD+DRwsUnQoYEB0yePjbw=="
+        }
+      ]
+    },
+    {
+      "id": "gridstack_js",
+      "kind": "npm_file",
+      "package": "gridstack",
+      "version": "12.4.2",
+      "license": "MIT",
+      "homepage": "http://gridstackjs.com/",
+      "official_url": "https://gridstackjs.com/",
+      "purpose": "GridStack bundle used by the dashboard",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/gridstack/-/gridstack-12.4.2.tgz",
+      "dist_integrity": "sha512-aXbJrQpi3LwpYXYOr4UriPM5uc/dPcjK01SdOE5PDpx2vi8tnLhU7yBg/1i4T59UhNkG/RBfabdFUObuN+gMnw==",
+      "source_path": "dist/gridstack-all.js",
+      "targets": [
+        {
+          "path": "core/static/js/gridstack-all.js",
+          "sha256": "e055ca4eb8bdc65f14f38c966ec31960c4c8dc6dc42e91f421bd629808185518",
+          "sri_sha512": "sha512-djBPxwvBhDep1SvOhliatweHMORhVO3HabrfBjaW6nYsa7UcJYHty31x42m4HBSJXcJSQdoEgRPLVYGGIuIaDQ=="
+        }
+      ]
+    },
+    {
+      "id": "jquery_js",
+      "kind": "npm_file",
+      "package": "jquery",
+      "version": "3.7.1",
+      "license": "MIT",
+      "homepage": "https://jquery.com",
+      "official_url": "https://jquery.com",
+      "purpose": "Latest jQuery 3.x release for compatibility with legacy plugins",
+      "notes": "The latest npm release is jQuery 4.x, but this project still vendors 3.7.1 to avoid breaking older plugins.",
+      "resolved": "https://registry.npmjs.org/jquery/-/jquery-3.7.1.tgz",
+      "dist_integrity": "sha512-m4avr8yL8kmFN8psrbFFFmB/If14iN5o9nw/NgnnM+kybDJpRsAynV2BsfpTYrTRysYUdADVD7CkUUizgkpLfg==",
+      "source_path": "dist/jquery.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/jquery.min.js",
+          "sha256": "fc9a93dd241f6b045cbff0481cf4e1901becd0e12fb45166a8f17f95823f0b1a",
+          "sri_sha512": "sha512-v2CJ7UaYy4JwqLDIrZUI/4hqeoQieOmAZNXBeQyjo21dadnwR+8ZaIJVT8EE2iyI61OV8e6M8PP2/4hpQINQ/g=="
+        }
+      ]
+    },
+    {
+      "id": "htmx_js",
+      "kind": "npm_file",
+      "package": "htmx.org",
+      "version": "2.0.8",
+      "license": "0BSD",
+      "homepage": "https://htmx.org/",
+      "official_url": "https://htmx.org/",
+      "purpose": "htmx runtime",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/htmx.org/-/htmx.org-2.0.8.tgz",
+      "dist_integrity": "sha512-fm297iru0iWsNJlBrjvtN7V9zjaxd+69Oqjh4F/Vq9Wwi2kFisLcrLCiv5oBX0KLfOX/zG8AUo9ROMU5XUB44Q==",
+      "source_path": "dist/htmx.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/htmx.min.js",
+          "sha256": "22283ef68cb7545914f0a88a1bdedc7256a703d1d580c1d255217d0a50d31313",
+          "sri_sha512": "sha512-CGXFnDNv5q48ciFeIyWFcfZhqYW0sSBiPO+HZDO3XLM+p8xjhezz5CCxtkXVDKfCbvF+iUhel7xoeSp19o7x7g=="
+        }
+      ]
+    },
+    {
+      "id": "hyperscript_js",
+      "kind": "npm_file",
+      "package": "hyperscript.org",
+      "version": "0.9.14",
+      "license": "BSD 2-Clause",
+      "homepage": "https://hyperscript.org/",
+      "official_url": "https://hyperscript.org/",
+      "purpose": "_hyperscript runtime",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/hyperscript.org/-/hyperscript.org-0.9.14.tgz",
+      "dist_integrity": "sha512-ugmojsQQUMmXcnwaXYiYf8L3GbeANy/m59EmE/0Z6C5eQ52fOuSrvFkuEIejG9BdpbYB4iTtoYGqV99eYqDVMA==",
+      "source_path": "dist/_hyperscript.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/hyperscript.min.js",
+          "sha256": "3e834a3ffc0334fee54ecff4e37a6ae951cd83e6daa96651ca7cfd8f751ad4d2",
+          "sri_sha512": "sha512-l43sZzpnAddmYhJyfPrgv46XhJvA95gsA28/+eW4XZLSekQ8wlP68i9f22KGkRjY0HNiZrLc5MXGo4z/tM2QNA=="
+        }
+      ]
+    },
+    {
+      "id": "magnet_js",
+      "kind": "npm_file",
+      "package": "@lf2com/magnet.js",
+      "version": "2.0.1",
+      "license": "MIT",
+      "homepage": "https://github.com/lf2com/magnet.js",
+      "official_url": "https://github.com/lf2com/magnet.js",
+      "purpose": "Magnet.js drag attraction component",
+      "notes": "",
+      "resolved": "https://registry.npmjs.org/@lf2com/magnet.js/-/magnet.js-2.0.1.tgz",
+      "dist_integrity": "sha512-MDgv1s0aNOuftuhY9c9Ve6Yadkmn7G+Ww91cVciyHHMhPPdxTxX3XUSJXFYD3VraGFzcnI4uilik9/I76AsJEg==",
+      "source_path": "dist/magnet.min.js",
+      "targets": [
+        {
+          "path": "core/static/js/magnet.min.js",
+          "sha256": "05ff8858b5fb7b3ad2a618212571162e2108f580b08527716f4e63c648dcccb1",
+          "sri_sha512": "sha512-aoQ3V4iCM8zTcdMDSUTRG1K9wqZzmDSisuaCLQexk9DdFy92oWvTUoAfCVLnGzzJClst8PmtasZg219REwyNkw=="
+        }
+      ]
+    },
+    {
+      "id": "fontawesome_bundle",
+      "kind": "url_bundle",
+      "version": "6.1.1",
+      "license": "Font Awesome Pro commercial terms via site-assets.fontawesome.com",
+      "homepage": "https://fontawesome.com",
+      "official_url": "https://fontawesome.com",
+      "purpose": "Existing Font Awesome asset family, self-hosted locally to preserve the currently used icon set",
+      "notes": "",
+      "entry_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/all.css",
+      "downloaded_files": [
+        {
+          "relative_path": "css/all.css",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/all.css"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-brands-400.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-brands-400.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-brands-400.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-brands-400.ttf"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-duotone-900.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-duotone-900.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-duotone-900.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-duotone-900.ttf"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-light-300.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-light-300.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-light-300.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-light-300.ttf"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-regular-400.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-regular-400.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-regular-400.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-regular-400.ttf"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-solid-900.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-solid-900.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-solid-900.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-solid-900.ttf"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-thin-100.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-thin-100.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-thin-100.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-thin-100.ttf"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-v4compatibility.woff2",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-v4compatibility.woff2"
+        },
+        {
+          "relative_path": "css/../webfonts/fa-v4compatibility.ttf",
+          "download_url": "https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-v4compatibility.ttf"
+        }
+      ],
+      "targets": [
+        {
+          "path": "core/static/vendor/fontawesome/css/all.css",
+          "sha256": "a35f901d01118e5649091bd03ac5784a7db52e111fb3806524c412f3d1dcfc5d",
+          "sri_sha512": "sha512-UKBBxJ5N3/MYiSsYTlEsARsp4vELKVRIklED4Mb6wpuVFOgy5Blt+sXUdz1TDReqWsm64xxBA2QoBJRCxI0x5Q=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-brands-400.woff2",
+          "sha256": "3701cbff3acccd80b1f2eede4311050514f7a64c2039eb77a77368fcd6e3de28",
+          "sri_sha512": "sha512-50+1yWldN03if6k/4jyncfSZyT44ev20Q7jmIGEiKG7v2qeB1nBOcVF9HD0mjSvAxmpS3+RDzoPfqbB4GfCkJg=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-brands-400.ttf",
+          "sha256": "06f6dc20f535066f0e069105329ad69d526d7a16215e7ff8af3dbc2cd0894186",
+          "sri_sha512": "sha512-klNly6qz+BwYPusuoZ0BIE01WW0CbrJszi5GrgZ6yFXQCrxAQRus6vMMATIxwPNYOYXu+UOd2KbRAsenhIkAGQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-duotone-900.woff2",
+          "sha256": "6f28dce91f45bc4687582137bb5d82d9771efc774e3b2b83c30018469d191ad8",
+          "sri_sha512": "sha512-Sz0g1G5m7symrIy2LfJOdASOrJQUo+IDIMpLKhFKm4RM7MLjV6wqlVGtDf2lEqE/tSzU5LQTtsc3djfZwDXCFQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-duotone-900.ttf",
+          "sha256": "6687687c70d502ff006750978f55d6d5108c364eec78db9b845adcfc838b78a4",
+          "sri_sha512": "sha512-1301N0n8/h7Kjp6E7LbEe7LWCyDPkfvYn9COlQ6ISbQTP/uPThs8NLhSr+nQskAPzjDoMqfojg0EyNfsD84s/w=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-light-300.woff2",
+          "sha256": "515954fe1dc163277d36b51f79fe56265f6b6cf79f99e307bbf6e52b477b9c87",
+          "sri_sha512": "sha512-xdrnQ7rYHIz80KJgGizTu80jCcWF4tGt/inACAoWT3dl3BleiIjq/g90RA42wJNcLpz3n8JAM1Z0ayUGROP5RQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-light-300.ttf",
+          "sha256": "7c41f14e1e8bbe7780049512c631b5301936a985dc6bbadd74a0cbc05549769c",
+          "sri_sha512": "sha512-i3tSu80Wfsx/12vQQziHhYQLjBNYLGQZfeXfyurZhCBibO+lq6kX1zARiktltXxBszu513BIKbnfzLTRYamoMQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-regular-400.woff2",
+          "sha256": "121b176974226dbc9b1ab227becb657d40b88d2bb7010a746c2360c31d7c373e",
+          "sri_sha512": "sha512-qioT43fXB5q4Bbpn8sPQE9OIZLjKD0c0lVmpm6KmT8k34LM6gkRcOOMi1BOl2lohFG/7p9tzKfTP5G563BQq1g=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-regular-400.ttf",
+          "sha256": "daa07950214eae5603ebb5a582a694da7b31a4c93f3bf38e9f616122860d83a5",
+          "sri_sha512": "sha512-1Ttr+eWt+Z7Na3gRZklbYEnBYGbQUATusZ4oeYLamI4iIjN8ILNV/upwk/h8qIJ2lf/Mukkj5zBVhUdXWJWLwQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-solid-900.woff2",
+          "sha256": "f350c708b5e7748a452b4b98600fa49127166d995686e260ccafb58d51a4ea62",
+          "sri_sha512": "sha512-Ph1xTLhfMycYSW+wUN8oL3Ggl56nGIS95EHiKWggcL/GbMNjPdib1Hreb1D4COlMxdiGCkk43nspQnpDuTjgQg=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-solid-900.ttf",
+          "sha256": "bbab2beb558f176d68d3e6002e4ea608633f7e6347dc6245dc67f8ad1c9ca18a",
+          "sri_sha512": "sha512-c/IBq1JAYcB6z9QnrVqSJREvhrp3NudGg8phmemZ1P41RTIc/z2txqbzP/dWVZYzZl9mPxDnDNLGTFWIQsAfPQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-thin-100.woff2",
+          "sha256": "92fb7777eb1a6a9c8e94048403db3e197e5e541bfd8142255e74ac69141081b2",
+          "sri_sha512": "sha512-U61hRcHKBdwKMsPzon6yDsO7ReY5gCHX3uztO+D48ynrqy4ZESWdoBtKohmJuXxry3epUhLQLsMd62nb6zmzhA=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-thin-100.ttf",
+          "sha256": "b8ef1ff99503a5611531d09bf5815261c1fb2f8291ae562711a7d7b651f92b9f",
+          "sri_sha512": "sha512-kFuJup+kEGELeelH6CXGKPaQ+cJfpLbOtbFFae6CwZfAidwsoM74wYu1Kzwb+pgeK9ODZc7SQMXpFWj5Mn7KtQ=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-v4compatibility.woff2",
+          "sha256": "4468a4ec439fecc93d86936bc3b5b5db7d0b98ce173faefd18938b2cef79245b",
+          "sri_sha512": "sha512-DzAaC0v4WatRDbN7LVoGXD5pmuOmveGoBimQLfnQFz74+L+hhZVatKhENPHAVkeNtRTBQ63B8w0fOlrQ3bERuw=="
+        },
+        {
+          "path": "core/static/vendor/fontawesome/css/../webfonts/fa-v4compatibility.ttf",
+          "sha256": "0f40286f817bb931f65fc1a5963f1450af855547fa9d2c02a009ad389fcc7d95",
+          "sri_sha512": "sha512-emfG/sIso1lhi9tNNmqIGP2xNU8SP47qPspS7g8V7ltbotbznl/wy00DxyjrtfxhYw3OfYYlxtyiGyoyRAFXTg=="
+        }
+      ]
+    }
+  ]
+}
--- a/artifacts/frontend_libraries.md
+++ b/artifacts/frontend_libraries.md
@@ -0,0 +1,315 @@
+# Frontend Library Inventory
+
+This report is generated by `scripts/vendor_frontend_assets.py` from `tools/frontend_assets/asset-manifest.json`.
+
+## bulma_css
+
+- Source: `bulma`
+- Version: `1.0.4`
+- Official URL: https://bulma.io
+- Homepage: https://bulma.io
+- License: `MIT`
+- Purpose: Bulma core stylesheet
+- Resolved tarball: `https://registry.npmjs.org/bulma/-/bulma-1.0.4.tgz`
+- Upstream package integrity: `sha512-Ffb6YGXDiZYX3cqvSbHWqQ8+LkX6tVoTcZuVB3lm93sbAVXlO0D6QlOTMnV6g18gILpAXqkG2z9hf9z4hCjz2g==`
+- Local targets:
+  - `core/static/css/bulma.min.css`
+    - SHA-256: `67fa26df1ca9e95d8f2adc7c04fa1b15fa3d24257470ebc10cc68b9aab914bee`
+    - SRI sha512: `sha512-yh2RE0wZCVZeysGiqTwDTO/dKelCbS9bP2L94UvOFtl/FKXcNAje3Y2oBg/ZMZ3LS1sicYk4dYVGtDex75fvvA==`
+
+
+## bulma_tooltip_css
+
+- Source: `bulma-tooltip`
+- Version: `3.0.2`
+- Official URL: https://wikiki.github.io/elements/tooltip
+- Homepage: https://github.com/Wikiki/bulma-tooltip#readme
+- License: `MIT`
+- Purpose: Official Bulma tooltip extension from Bulma's extensions page
+- Resolved tarball: `https://registry.npmjs.org/bulma-tooltip/-/bulma-tooltip-3.0.2.tgz`
+- Upstream package integrity: `sha512-CsT3APjhlZScskFg38n8HYL8oYNUHQtcu4sz6ERarxkUpBRbk9v0h/5KAvXeKapVSn2dp9l7bOGit5SECP8EWQ==`
+- Local targets:
+  - `core/static/css/bulma-tooltip.min.css`
+    - SHA-256: `5c79d12a40b3532aaec159faa0b85fd3d500e192467761b71e0bda0fd04f3076`
+    - SRI sha512: `sha512-SNDNIUvSYhnqDV9FFXaH/e0xZ6NzkG4Qm5dafLLf0PCMkzICKaOmMTgI3y2t2jZK+hAtP6A7UBcFqjWMhsujIg==`
+
+
+## bulma_slider_css
+
+- Source: `bulma-slider`
+- Version: `2.0.5`
+- Official URL: https://wikiki.github.io/form/slider
+- Homepage: https://github.com/Wikiki/bulma-slider#readme
+- License: `MIT`
+- Purpose: Official Bulma slider extension from Bulma's extensions page
+- Resolved tarball: `https://registry.npmjs.org/bulma-slider/-/bulma-slider-2.0.5.tgz`
+- Upstream package integrity: `sha512-6woD/1E7q1o5bfEaQjNqpWZaCItC1oHe9bN15WYB2ELqz2gDaJYZkf+rlozGpAYOXQGDQGCCv3y+QuKjx6sQuw==`
+- Local targets:
+  - `core/static/css/bulma-slider.min.css`
+    - SHA-256: `f9d952627d388b8ba267e1388d6923274cf9e62e758d459c5a045f3933e9dc8a`
+    - SRI sha512: `sha512-9o5SkCRCA9thttRH3Gb5QXLxKdRiuRLdO6ToEPwRHGLXjrhTZwFj0rEHjrCcJvDN9/aNaWMpGOIEA2vZsHmEqw==`
+
+
+## bulma_slider_js
+
+- Source: `bulma-slider`
+- Version: `2.0.5`
+- Official URL: https://wikiki.github.io/form/slider
+- Homepage: https://github.com/Wikiki/bulma-slider#readme
+- License: `MIT`
+- Purpose: Official Bulma slider extension runtime
+- Resolved tarball: `https://registry.npmjs.org/bulma-slider/-/bulma-slider-2.0.5.tgz`
+- Upstream package integrity: `sha512-6woD/1E7q1o5bfEaQjNqpWZaCItC1oHe9bN15WYB2ELqz2gDaJYZkf+rlozGpAYOXQGDQGCCv3y+QuKjx6sQuw==`
+- Local targets:
+  - `core/static/js/bulma-slider.min.js`
+    - SHA-256: `db68ebe154a25597913c5635f31500fe7a32e5a205fb9a98c9642d0c2de47d9e`
+    - SRI sha512: `sha512-WLKXHCsMXTSIPsmQShJRE6K4IzwvNkhwxr/Oo8N3z+kzjhGleHibspmWLTawNMdl2z9E23XK20+yvUTDZ+zeNQ==`
+
+
+## bulma_calendar_css
+
+- Source: `bulma-calendar`
+- Version: `7.1.1`
+- Official URL: https://wikiki.github.io/components/calendar
+- Homepage: https://doc.mh-s.de/bulma-calendar
+- License: `MIT`
+- Purpose: Official Bulma calendar extension from Bulma's extensions page
+- Resolved tarball: `https://registry.npmjs.org/bulma-calendar/-/bulma-calendar-7.1.1.tgz`
+- Upstream package integrity: `sha512-E08i25KOfqMKBndgDF3y3eoQ0dUzVkgV9R53EDRM65GQUQKLzt8gcXVJYs3mYnpq6L3DiLuUt47Fl09tSv9OpA==`
+- Local targets:
+  - `core/static/css/bulma-calendar.min.css`
+    - SHA-256: `d18b488ca52584bcd6ea3fb84bf06380e47a3cd18660a235617da017d13ab269`
+    - SRI sha512: `sha512-IOnJQkgQpezPDPTJcRiWD7YVI3sF2RYzYDl4isbDT2geSaEHRQ615UN/8GhJbSkvqkKRZu8SBCQ7XwKMqsqLFQ==`
+
+
+## bulma_calendar_js
+
+- Source: `bulma-calendar`
+- Version: `7.1.1`
+- Official URL: https://wikiki.github.io/components/calendar
+- Homepage: https://doc.mh-s.de/bulma-calendar
+- License: `MIT`
+- Purpose: Official Bulma calendar extension runtime
+- Resolved tarball: `https://registry.npmjs.org/bulma-calendar/-/bulma-calendar-7.1.1.tgz`
+- Upstream package integrity: `sha512-E08i25KOfqMKBndgDF3y3eoQ0dUzVkgV9R53EDRM65GQUQKLzt8gcXVJYs3mYnpq6L3DiLuUt47Fl09tSv9OpA==`
+- Local targets:
+  - `core/static/js/bulma-calendar.min.js`
+    - SHA-256: `58160c87c4d17f9d98ec366fe019492acde50efbc0297af7045547952b306680`
+    - SRI sha512: `sha512-kkEtEtypXzruevjkoxhyEkqkZBtlhK7s8zt7IV2yPabgBwy5xbKL9uWeCS37ldS9AaNTSnveWTu4ivUvGMJUWA==`
+
+
+## bulma_tagsinput_css
+
+- Source: `bulma-tagsinput`
+- Version: `2.0.0`
+- Official URL: https://wikiki.github.io/form/tagsinput
+- Homepage: https://github.com/Wikiki/bulma-tagsinput#readme
+- License: `MIT`
+- Purpose: Official Bulma tagsinput extension from Bulma's extensions page
+- Resolved tarball: `https://registry.npmjs.org/bulma-tagsinput/-/bulma-tagsinput-2.0.0.tgz`
+- Upstream package integrity: `sha512-BFvd0oaxgeWHOEh3d4cgETy5vpSSjRRBA9w+8TWEuhjFQg38Rb+3vjDCavL+udpdjf+dRV0SK5T4kYCXTOrz5A==`
+- Local targets:
+  - `core/static/css/bulma-tagsinput.min.css`
+    - SHA-256: `8d1de24619c05ddf9045638b52059ab492d4887ce74119eed545d66af859da89`
+    - SRI sha512: `sha512-NWTkcDRubZ3pyXbZZLQBILuVsRFs8c6QGgnfe4dm5/d6yp50U+xdoCDLIcSo51fFy/GXH0O2Oed1Z1sF1faxDA==`
+
+
+## bulma_tagsinput_js
+
+- Source: `bulma-tagsinput`
+- Version: `2.0.0`
+- Official URL: https://wikiki.github.io/form/tagsinput
+- Homepage: https://github.com/Wikiki/bulma-tagsinput#readme
+- License: `MIT`
+- Purpose: Official Bulma tagsinput extension runtime
+- Resolved tarball: `https://registry.npmjs.org/bulma-tagsinput/-/bulma-tagsinput-2.0.0.tgz`
+- Upstream package integrity: `sha512-BFvd0oaxgeWHOEh3d4cgETy5vpSSjRRBA9w+8TWEuhjFQg38Rb+3vjDCavL+udpdjf+dRV0SK5T4kYCXTOrz5A==`
+- Local targets:
+  - `core/static/js/bulma-tagsinput.min.js`
+    - SHA-256: `b355aa94ec519e374d7edf569e3dbde8bbe30ff3a193cb96f2930ee7815939d6`
+    - SRI sha512: `sha512-Je6J++MjmmpxF30JCmRwM2KiK3uWQBQtqiNCjwzEMJKExLaa0BqerlYNa/fJAl5Rra4hMgRZF2fzg+V2vjE4Kw==`
+
+
+## bulma_switch_css
+
+- Source: `bulma-switch`
+- Version: `2.0.4`
+- Official URL: https://wikiki.github.io/form/switch
+- Homepage: https://github.com/Wikiki/bulma-switch#readme
+- License: `MIT`
+- Purpose: Official Bulma switch extension from Bulma's extensions page
+- Resolved tarball: `https://registry.npmjs.org/bulma-switch/-/bulma-switch-2.0.4.tgz`
+- Upstream package integrity: `sha512-kMu4H0Pr0VjvfsnT6viRDCgptUq0Rvy7y7PX6q+IHg1xUynsjszPjhAdal5ysAlCG5HNO+5YXxeiu92qYGQolw==`
+- Local targets:
+  - `core/static/css/bulma-switch.min.css`
+    - SHA-256: `f0460ddebdd95425a50590908503a170f5ff08b28bd53573c71791fc7cd1e6f5`
+    - SRI sha512: `sha512-zjrHYubQoNgDVqVKTyGjKcvIeQlduZTvXCvcBwQ0iqJYKLKiz9cuFAN7e98zfKqCTpI/EgFRBRcTwJw20yAFuw==`
+
+
+## gridstack_css
+
+- Source: `gridstack`
+- Version: `12.4.2`
+- Official URL: https://gridstackjs.com/
+- Homepage: http://gridstackjs.com/
+- License: `MIT`
+- Purpose: GridStack stylesheet
+- Resolved tarball: `https://registry.npmjs.org/gridstack/-/gridstack-12.4.2.tgz`
+- Upstream package integrity: `sha512-aXbJrQpi3LwpYXYOr4UriPM5uc/dPcjK01SdOE5PDpx2vi8tnLhU7yBg/1i4T59UhNkG/RBfabdFUObuN+gMnw==`
+- Local targets:
+  - `core/static/css/gridstack.min.css`
+    - SHA-256: `55e9d4ea6d8c6f8f1ea8a449b8af18d8571487c0afc6b433cccf877047cb8457`
+    - SRI sha512: `sha512-ttQfsDTO64bamkJHeLDf0kzMP1NKfkootudPWS2V8Pwy+9z1wexSYjIT6/HXGg/bmtD+DRwsUnQoYEB0yePjbw==`
+
+
+## gridstack_js
+
+- Source: `gridstack`
+- Version: `12.4.2`
+- Official URL: https://gridstackjs.com/
+- Homepage: http://gridstackjs.com/
+- License: `MIT`
+- Purpose: GridStack bundle used by the dashboard
+- Resolved tarball: `https://registry.npmjs.org/gridstack/-/gridstack-12.4.2.tgz`
+- Upstream package integrity: `sha512-aXbJrQpi3LwpYXYOr4UriPM5uc/dPcjK01SdOE5PDpx2vi8tnLhU7yBg/1i4T59UhNkG/RBfabdFUObuN+gMnw==`
+- Local targets:
+  - `core/static/js/gridstack-all.js`
+    - SHA-256: `e055ca4eb8bdc65f14f38c966ec31960c4c8dc6dc42e91f421bd629808185518`
+    - SRI sha512: `sha512-djBPxwvBhDep1SvOhliatweHMORhVO3HabrfBjaW6nYsa7UcJYHty31x42m4HBSJXcJSQdoEgRPLVYGGIuIaDQ==`
+
+
+## jquery_js
+
+- Source: `jquery`
+- Version: `3.7.1`
+- Official URL: https://jquery.com
+- Homepage: https://jquery.com
+- License: `MIT`
+- Purpose: Latest jQuery 3.x release for compatibility with legacy plugins
+- Resolved tarball: `https://registry.npmjs.org/jquery/-/jquery-3.7.1.tgz`
+- Upstream package integrity: `sha512-m4avr8yL8kmFN8psrbFFFmB/If14iN5o9nw/NgnnM+kybDJpRsAynV2BsfpTYrTRysYUdADVD7CkUUizgkpLfg==`
+- Notes: The latest npm release is jQuery 4.x, but this project still vendors 3.7.1 to avoid breaking older plugins.
+- Local targets:
+  - `core/static/js/jquery.min.js`
+    - SHA-256: `fc9a93dd241f6b045cbff0481cf4e1901becd0e12fb45166a8f17f95823f0b1a`
+    - SRI sha512: `sha512-v2CJ7UaYy4JwqLDIrZUI/4hqeoQieOmAZNXBeQyjo21dadnwR+8ZaIJVT8EE2iyI61OV8e6M8PP2/4hpQINQ/g==`
+
+
+## htmx_js
+
+- Source: `htmx.org`
+- Version: `2.0.8`
+- Official URL: https://htmx.org/
+- Homepage: https://htmx.org/
+- License: `0BSD`
+- Purpose: htmx runtime
+- Resolved tarball: `https://registry.npmjs.org/htmx.org/-/htmx.org-2.0.8.tgz`
+- Upstream package integrity: `sha512-fm297iru0iWsNJlBrjvtN7V9zjaxd+69Oqjh4F/Vq9Wwi2kFisLcrLCiv5oBX0KLfOX/zG8AUo9ROMU5XUB44Q==`
+- Local targets:
+  - `core/static/js/htmx.min.js`
+    - SHA-256: `22283ef68cb7545914f0a88a1bdedc7256a703d1d580c1d255217d0a50d31313`
+    - SRI sha512: `sha512-CGXFnDNv5q48ciFeIyWFcfZhqYW0sSBiPO+HZDO3XLM+p8xjhezz5CCxtkXVDKfCbvF+iUhel7xoeSp19o7x7g==`
+
+
+## hyperscript_js
+
+- Source: `hyperscript.org`
+- Version: `0.9.14`
+- Official URL: https://hyperscript.org/
+- Homepage: https://hyperscript.org/
+- License: `BSD 2-Clause`
+- Purpose: _hyperscript runtime
+- Resolved tarball: `https://registry.npmjs.org/hyperscript.org/-/hyperscript.org-0.9.14.tgz`
+- Upstream package integrity: `sha512-ugmojsQQUMmXcnwaXYiYf8L3GbeANy/m59EmE/0Z6C5eQ52fOuSrvFkuEIejG9BdpbYB4iTtoYGqV99eYqDVMA==`
+- Local targets:
+  - `core/static/js/hyperscript.min.js`
+    - SHA-256: `3e834a3ffc0334fee54ecff4e37a6ae951cd83e6daa96651ca7cfd8f751ad4d2`
+    - SRI sha512: `sha512-l43sZzpnAddmYhJyfPrgv46XhJvA95gsA28/+eW4XZLSekQ8wlP68i9f22KGkRjY0HNiZrLc5MXGo4z/tM2QNA==`
+
+
+## magnet_js
+
+- Source: `@lf2com/magnet.js`
+- Version: `2.0.1`
+- Official URL: https://github.com/lf2com/magnet.js
+- Homepage: https://github.com/lf2com/magnet.js
+- License: `MIT`
+- Purpose: Magnet.js drag attraction component
+- Resolved tarball: `https://registry.npmjs.org/@lf2com/magnet.js/-/magnet.js-2.0.1.tgz`
+- Upstream package integrity: `sha512-MDgv1s0aNOuftuhY9c9Ve6Yadkmn7G+Ww91cVciyHHMhPPdxTxX3XUSJXFYD3VraGFzcnI4uilik9/I76AsJEg==`
+- Local targets:
+  - `core/static/js/magnet.min.js`
+    - SHA-256: `05ff8858b5fb7b3ad2a618212571162e2108f580b08527716f4e63c648dcccb1`
+    - SRI sha512: `sha512-aoQ3V4iCM8zTcdMDSUTRG1K9wqZzmDSisuaCLQexk9DdFy92oWvTUoAfCVLnGzzJClst8PmtasZg219REwyNkw==`
+
+
+## fontawesome_bundle
+
+- Source: `https://site-assets.fontawesome.com/releases/v6.1.1/css/all.css`
+- Version: `6.1.1`
+- Official URL: https://fontawesome.com
+- Homepage: https://fontawesome.com
+- License: `Font Awesome Pro commercial terms via site-assets.fontawesome.com`
+- Purpose: Existing Font Awesome asset family, self-hosted locally to preserve the currently used icon set
+- Downloaded bundle files:
+  - `css/all.css` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/all.css`
+  - `css/../webfonts/fa-brands-400.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-brands-400.woff2`
+  - `css/../webfonts/fa-brands-400.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-brands-400.ttf`
+  - `css/../webfonts/fa-duotone-900.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-duotone-900.woff2`
+  - `css/../webfonts/fa-duotone-900.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-duotone-900.ttf`
+  - `css/../webfonts/fa-light-300.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-light-300.woff2`
+  - `css/../webfonts/fa-light-300.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-light-300.ttf`
+  - `css/../webfonts/fa-regular-400.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-regular-400.woff2`
+  - `css/../webfonts/fa-regular-400.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-regular-400.ttf`
+  - `css/../webfonts/fa-solid-900.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-solid-900.woff2`
+  - `css/../webfonts/fa-solid-900.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-solid-900.ttf`
+  - `css/../webfonts/fa-thin-100.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-thin-100.woff2`
+  - `css/../webfonts/fa-thin-100.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-thin-100.ttf`
+  - `css/../webfonts/fa-v4compatibility.woff2` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-v4compatibility.woff2`
+  - `css/../webfonts/fa-v4compatibility.ttf` from `https://site-assets.fontawesome.com/releases/v6.1.1/css/webfonts/fa-v4compatibility.ttf`
+- Local targets:
+  - `core/static/vendor/fontawesome/css/all.css`
+    - SHA-256: `a35f901d01118e5649091bd03ac5784a7db52e111fb3806524c412f3d1dcfc5d`
+    - SRI sha512: `sha512-UKBBxJ5N3/MYiSsYTlEsARsp4vELKVRIklED4Mb6wpuVFOgy5Blt+sXUdz1TDReqWsm64xxBA2QoBJRCxI0x5Q==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-brands-400.woff2`
+    - SHA-256: `3701cbff3acccd80b1f2eede4311050514f7a64c2039eb77a77368fcd6e3de28`
+    - SRI sha512: `sha512-50+1yWldN03if6k/4jyncfSZyT44ev20Q7jmIGEiKG7v2qeB1nBOcVF9HD0mjSvAxmpS3+RDzoPfqbB4GfCkJg==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-brands-400.ttf`
+    - SHA-256: `06f6dc20f535066f0e069105329ad69d526d7a16215e7ff8af3dbc2cd0894186`
+    - SRI sha512: `sha512-klNly6qz+BwYPusuoZ0BIE01WW0CbrJszi5GrgZ6yFXQCrxAQRus6vMMATIxwPNYOYXu+UOd2KbRAsenhIkAGQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-duotone-900.woff2`
+    - SHA-256: `6f28dce91f45bc4687582137bb5d82d9771efc774e3b2b83c30018469d191ad8`
+    - SRI sha512: `sha512-Sz0g1G5m7symrIy2LfJOdASOrJQUo+IDIMpLKhFKm4RM7MLjV6wqlVGtDf2lEqE/tSzU5LQTtsc3djfZwDXCFQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-duotone-900.ttf`
+    - SHA-256: `6687687c70d502ff006750978f55d6d5108c364eec78db9b845adcfc838b78a4`
+    - SRI sha512: `sha512-1301N0n8/h7Kjp6E7LbEe7LWCyDPkfvYn9COlQ6ISbQTP/uPThs8NLhSr+nQskAPzjDoMqfojg0EyNfsD84s/w==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-light-300.woff2`
+    - SHA-256: `515954fe1dc163277d36b51f79fe56265f6b6cf79f99e307bbf6e52b477b9c87`
+    - SRI sha512: `sha512-xdrnQ7rYHIz80KJgGizTu80jCcWF4tGt/inACAoWT3dl3BleiIjq/g90RA42wJNcLpz3n8JAM1Z0ayUGROP5RQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-light-300.ttf`
+    - SHA-256: `7c41f14e1e8bbe7780049512c631b5301936a985dc6bbadd74a0cbc05549769c`
+    - SRI sha512: `sha512-i3tSu80Wfsx/12vQQziHhYQLjBNYLGQZfeXfyurZhCBibO+lq6kX1zARiktltXxBszu513BIKbnfzLTRYamoMQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-regular-400.woff2`
+    - SHA-256: `121b176974226dbc9b1ab227becb657d40b88d2bb7010a746c2360c31d7c373e`
+    - SRI sha512: `sha512-qioT43fXB5q4Bbpn8sPQE9OIZLjKD0c0lVmpm6KmT8k34LM6gkRcOOMi1BOl2lohFG/7p9tzKfTP5G563BQq1g==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-regular-400.ttf`
+    - SHA-256: `daa07950214eae5603ebb5a582a694da7b31a4c93f3bf38e9f616122860d83a5`
+    - SRI sha512: `sha512-1Ttr+eWt+Z7Na3gRZklbYEnBYGbQUATusZ4oeYLamI4iIjN8ILNV/upwk/h8qIJ2lf/Mukkj5zBVhUdXWJWLwQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-solid-900.woff2`
+    - SHA-256: `f350c708b5e7748a452b4b98600fa49127166d995686e260ccafb58d51a4ea62`
+    - SRI sha512: `sha512-Ph1xTLhfMycYSW+wUN8oL3Ggl56nGIS95EHiKWggcL/GbMNjPdib1Hreb1D4COlMxdiGCkk43nspQnpDuTjgQg==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-solid-900.ttf`
+    - SHA-256: `bbab2beb558f176d68d3e6002e4ea608633f7e6347dc6245dc67f8ad1c9ca18a`
+    - SRI sha512: `sha512-c/IBq1JAYcB6z9QnrVqSJREvhrp3NudGg8phmemZ1P41RTIc/z2txqbzP/dWVZYzZl9mPxDnDNLGTFWIQsAfPQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-thin-100.woff2`
+    - SHA-256: `92fb7777eb1a6a9c8e94048403db3e197e5e541bfd8142255e74ac69141081b2`
+    - SRI sha512: `sha512-U61hRcHKBdwKMsPzon6yDsO7ReY5gCHX3uztO+D48ynrqy4ZESWdoBtKohmJuXxry3epUhLQLsMd62nb6zmzhA==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-thin-100.ttf`
+    - SHA-256: `b8ef1ff99503a5611531d09bf5815261c1fb2f8291ae562711a7d7b651f92b9f`
+    - SRI sha512: `sha512-kFuJup+kEGELeelH6CXGKPaQ+cJfpLbOtbFFae6CwZfAidwsoM74wYu1Kzwb+pgeK9ODZc7SQMXpFWj5Mn7KtQ==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-v4compatibility.woff2`
+    - SHA-256: `4468a4ec439fecc93d86936bc3b5b5db7d0b98ce173faefd18938b2cef79245b`
+    - SRI sha512: `sha512-DzAaC0v4WatRDbN7LVoGXD5pmuOmveGoBimQLfnQFz74+L+hhZVatKhENPHAVkeNtRTBQ63B8w0fOlrQ3bERuw==`
+  - `core/static/vendor/fontawesome/css/../webfonts/fa-v4compatibility.ttf`
+    - SHA-256: `0f40286f817bb931f65fc1a5963f1450af855547fa9d2c02a009ad389fcc7d95`
+    - SRI sha512: `sha512-emfG/sIso1lhi9tNNmqIGP2xNU8SP47qPspS7g8V7ltbotbznl/wy00DxyjrtfxhYw3OfYYlxtyiGyoyRAFXTg==`
--- a/artifacts/mcp/manticore-mcp-server.md
+++ b/artifacts/mcp/manticore-mcp-server.md
@@ -0,0 +1,71 @@
+# Manticore MCP Server (GIA)
+
+This document describes the MCP server wired for task + memory operations in GIA.
+
+## Server entrypoint
+
+- Django management command: `python manage.py mcp_manticore_server`
+- Python module: `core.mcp.server`
+- Tool handlers: `core.mcp.tools`
+
+## Rust worker frontend (optional)
+
+For low-overhead direct Manticore operations, a Rust stdio MCP worker is included:
+
+- Project: `rust/manticore-mcp-worker`
+- Build: `make mcp-rust-build`
+- Binary: `rust/manticore-mcp-worker/target/release/manticore-mcp-worker`
+- VS Code server name: `manticore-rust-worker` (disabled by default in `/code/xf/.vscode/mcp.json`)
+
+This worker exposes fast table/status/query/maintenance operations and can be enabled when you want a minimal MCP process in front of Manticore.
+
+## VS Code wiring
+
+Workspace config is in `/code/xf/.vscode/mcp.json`:
+
+- Server name: `manticore`
+- Launch method: `podman exec -i ur_gia /venv/bin/python manage.py mcp_manticore_server`
+- Forced env:
+  - `MEMORY_SEARCH_BACKEND=manticore`
+
+`MANTICORE_HTTP_URL` is inherited from container environment so each deployment can set the correct reachable address.
+
+This allows MCP tool calls from VS Code to run against the live GIA container without requiring local Django dependencies.
+
+## Implemented MCP tools
+
+- `manticore.status`
+- `manticore.query`
+- `manticore.reindex`
+- `memory.list`
+- `memory.propose`
+- `memory.pending`
+- `memory.review`
+- `memory.suggest_from_messages`
+- `tasks.list`
+- `tasks.search`
+- `tasks.get`
+- `tasks.events`
+- `tasks.create_note`
+- `tasks.link_artifact`
+- `wiki.create_article`
+- `wiki.update_article`
+- `wiki.list`
+- `wiki.get`
+- `project.get_guidelines`
+- `project.get_layout`
+- `project.get_runbook`
+- `docs.append_run_note`
+
+`docs.append_run_note` appends markdown notes to `/tmp/gia-mcp-run-notes.md` by default (or a project path you pass explicitly).
+
+All MCP tool invocations are audit-logged in `core_mcptoolauditlog` (`MCPToolAuditLog` model).
+
+## Runtime notes
+
+1. Ensure GIA services are running (`make run`).
+2. Start Manticore container:
+   - `./utilities/memory/manage_manticore_container.sh up`
+3. Optional initial index:
+   - `podman exec ur_gia /venv/bin/python manage.py memory_search_reindex --user-id <id> --statuses active`
+4. In VS Code, approve/enabled the workspace MCP server when prompted.
--- a/artifacts/plans/03-media-asset-normalization.md
+++ b/artifacts/plans/03-media-asset-normalization.md
@@ -0,0 +1,28 @@
+# Feature Plan: Media Asset Normalization
+
+## Goal
+Normalize media into reusable assets so messages reference media IDs rather than duplicate transport blobs.
+
+## Why This Fits GIA
+- GIA already has shared media prep and cross-transport relay.
+- Normalized assets reduce duplicate downloads/uploads and improve traceability.
+
+## Scope
+- New `MediaAsset` entity with checksum + metadata.
+- New link table between `Message` and `MediaAsset`.
+- De-dup by `sha256` and file size.
+
+## Implementation
+1. Add models/migrations: `MediaAsset`, `MessageMediaRef`.
+2. Update inbound media ingestion to upsert `MediaAsset`.
+3. Update outbound transport prep to consume `MediaAsset` references.
+4. Add background backfill for existing attachments.
+5. Add retention/GC strategy for orphaned assets.
+
+## Acceptance Criteria
+- Same file across transports resolves to one `MediaAsset`.
+- Message records keep only pointers and display metadata.
+- Relay path can reuse existing asset without re-download.
+
+## Out of Scope
+- External CDN migration in this phase.
--- a/artifacts/plans/04-identity-resolution-and-merge.md
+++ b/artifacts/plans/04-identity-resolution-and-merge.md
@@ -0,0 +1,27 @@
+# Feature Plan: Identity Resolution and Merge
+
+## Goal
+Improve person graph quality by suggesting and applying safe merges across transport identifiers.
+
+## Why This Fits GIA
+- Core value of universal inbox depends on clean `Person` identity graph.
+
+## Scope
+- Heuristic scoring for candidate merges.
+- Manual review queue with approve/reject.
+- Merge operation with audit trail and undo window.
+
+## Implementation
+1. Add scoring service using signals: normalized phone, username similarity, name overlap, shared chat co-occurrence.
+2. Add `IdentityMergeSuggestion` model with score + reasons.
+3. Add UI panel to review/approve merges.
+4. Implement safe merge transaction (`PersonIdentifier` reassignment, metadata merge rules).
+5. Emit audit events and rollback snapshots.
+
+## Acceptance Criteria
+- Suggestions generated deterministically and explain reasons.
+- Merge is idempotent and reversible within configured window.
+- No identifier is lost or orphaned.
+
+## Out of Scope
+- Fully automatic merge without human approval.
--- a/artifacts/plans/05-adapter-resilience-supervisor.md
+++ b/artifacts/plans/05-adapter-resilience-supervisor.md
@@ -0,0 +1,27 @@
+# Feature Plan: Adapter Resilience Supervisor
+
+## Goal
+Make adapters self-healing and observable under disconnects, API drift, and transient faults.
+
+## Why This Fits GIA
+- Bridge reliability is a product requirement for multi-network ops.
+
+## Scope
+- Health probes per adapter.
+- Reconnect/backoff supervisor.
+- Circuit-breaker for repeated failure classes.
+
+## Implementation
+1. Add adapter health state model (healthy/degraded/down).
+2. Add watchdog jobs for Signal/WhatsApp/Instagram/XMPP.
+3. Implement exponential backoff + jitter reconnect policies.
+4. Emit structured adapter health events + alerts.
+5. Add status surface in settings/services pages.
+
+## Acceptance Criteria
+- Adapter restarts automatically on transient failures.
+- Repeated failures degrade state and stop spam retries.
+- Operators can view current health + last error + retry schedule.
+
+## Out of Scope
+- Full protocol hot-reload without process restart.
--- a/artifacts/plans/06-end-to-end-observability.md
+++ b/artifacts/plans/06-end-to-end-observability.md
@@ -0,0 +1,27 @@
+# Feature Plan: End-to-End Observability and Traceability
+
+## Goal
+Provide trace-level visibility from ingress transport event to UI delivery/ack.
+
+## Why This Fits GIA
+- Multi-hop messaging systems require correlation IDs to debug reliably.
+
+## Scope
+- Global trace IDs for message lifecycle.
+- Structured logs and timeline diagnostics view.
+- Basic metrics and SLA dashboards.
+
+## Implementation
+1. Inject `trace_id` at ingress/send initiation.
+2. Propagate through router, persistence, websocket, command/task flows.
+3. Standardize structured log schema across services.
+4. Add timeline diagnostics page by trace ID and session.
+5. Add core metrics: ingress latency, send latency, drop rate, retry counts.
+
+## Acceptance Criteria
+- One trace ID can reconstruct full message path.
+- At least 95% of critical paths emit structured trace logs.
+- Operators can isolate bottleneck stage in under 2 minutes.
+
+## Out of Scope
+- Full distributed tracing vendor integration.
--- a/artifacts/plans/07-unified-conversation-graph-ui.md
+++ b/artifacts/plans/07-unified-conversation-graph-ui.md
@@ -0,0 +1,25 @@
+# Feature Plan: Unified Conversation Graph UI
+
+## Goal
+Expose one-person-many-identifiers graph in UI so users work per person, not per transport fragment.
+
+## Why This Fits GIA
+- GIA already models `Person` + `PersonIdentifier`; this surfaces it clearly.
+
+## Scope
+- Person profile panel with linked identifiers and active sessions.
+- Cross-transport thread pivoting from one card.
+- Merge/split controls linked to identity suggestions.
+
+## Implementation
+1. Add graph view endpoint and serializer for person graph.
+2. Update compose sidebar/person pages with linked transport pills.
+3. Add quick actions: open thread, relink identifier, propose merge.
+4. Integrate with identity merge queue from feature 04.
+
+## Acceptance Criteria
+- User can see and navigate all transport identities for a person from one place.
+- Switching transport context preserves person-centric history access.
+
+## Out of Scope
+- Force-merged unified thread rendering across all transports.
--- a/artifacts/plans/08-conversation-intelligence-suite.md
+++ b/artifacts/plans/08-conversation-intelligence-suite.md
@@ -0,0 +1,27 @@
+# Feature Plan: Conversation Intelligence Suite
+
+## Goal
+Add actionable conversation analytics and AI-assisted summaries with transparent outputs.
+
+## Why This Fits GIA
+- AI workspace already exists; this extends it with communication-specific signals.
+
+## Scope
+- Metrics: response latency, engagement symmetry, activity trend.
+- AI summary for recent window with action item extraction.
+- Draft mediation suggestions with raw/original always visible.
+
+## Implementation
+1. Add analytics service computing per-thread/person metrics.
+2. Add summary generation endpoint with cached recent windows.
+3. Add action-item extraction that can feed tasks pipeline.
+4. Compose UI card: metrics + summary + suggested next reply.
+5. Add "show raw" indicator whenever mediation is displayed.
+
+## Acceptance Criteria
+- Metrics refresh reliably on new events.
+- Summaries are attributable to explicit message windows.
+- Mediation never hides original user text.
+
+## Out of Scope
+- Fully autonomous outbound sending.
--- a/artifacts/plans/10-policy-engine-middleware.md
+++ b/artifacts/plans/10-policy-engine-middleware.md
@@ -0,0 +1,27 @@
+# Feature Plan: Policy Engine Middleware
+
+## Goal
+Implement deterministic middleware policies for inbound/outbound message transforms and automations.
+
+## Why This Fits GIA
+- Existing command/task/approval flows provide natural control points.
+
+## Scope
+- Per-chat and per-person policy rules.
+- Rule actions: notify, suggest rewrite, create task, rate-limit send, require approval.
+- Dry-run/audit mode for every policy decision.
+
+## Implementation
+1. Add policy schema and evaluator (`if conditions -> actions`).
+2. Hook evaluator into compose send and inbound task intelligence paths.
+3. Add policy decision logs with trace IDs.
+4. Add UI for policy creation/testing with preview mode.
+5. Add safe defaults: all transform rules start as suggest-only.
+
+## Acceptance Criteria
+- Policy decisions are deterministic and replayable.
+- Any transform is visibly annotated in UI.
+- Users can disable policy per chat instantly.
+
+## Out of Scope
+- Natural-language-to-policy compiler in first release.
--- a/artifacts/plans/17-person-enrichment-without-llm.md
+++ b/artifacts/plans/17-person-enrichment-without-llm.md
@@ -0,0 +1,95 @@
+# Feature Plan: Person Model Enrichment (Non-LLM First)
+
+## Goal
+Populate `Person` fields from existing message history without spending OpenAI tokens by default:
+- `summary`
+- `profile`
+- `revealed`
+- `likes`
+- `dislikes`
+- `sentiment`
+- `timezone`
+- `last_interaction`
+
+## Problem We Are Solving
+- We have high-volume message data but limited durable person intelligence.
+- LLM analysis is expensive for continuous/background processing.
+- We need fast, deterministic extraction first, with optional semantic ranking.
+
+## Design Decisions
+1. Config scope:
+   - global defaults
+   - optional group-level overrides
+   - per-user overrides
+2. Resolution order:
+   - `user > group > global`
+3. Global toggle:
+   - hard kill-switch (`PERSON_ENRICHMENT_ENABLED`)
+4. Per-user/group controls:
+   - enable/disable enrichment
+   - write mode (`proposal_required` or `direct`)
+   - confidence threshold
+   - max messages scanned per run
+   - semantic-ranking toggle
+
+## Proposed Data Additions
+- `PersonEnrichmentSettings`:
+  - scope fields (`user`, optional `group`)
+  - toggle/threshold/runtime limits
+- `PersonSignal`:
+  - normalized extracted clue
+  - source references (message ids/events)
+  - confidence and detector name
+- `PersonUpdateProposal`:
+  - pending/approved/rejected person field updates
+  - reason and provenance
+- Optional `PersonFieldRevision`:
+  - before/after snapshots for auditability
+
+## Processing Flow
+1. Select message window:
+   - recent inbound/outbound messages per person/service
+   - bounded by configurable caps
+2. Fast extraction:
+   - deterministic rules/regex for:
+     - timezone cues
+     - explicit likes/dislikes
+     - self-revealed facts
+     - interaction-derived sentiment hints
+3. Semantic ranking (optional):
+   - use Manticore-backed similarity search for classifier labels
+   - rank candidate signals; do not call OpenAI in default path
+4. Signal aggregation:
+   - merge repeated evidence
+   - decay stale evidence
+   - detect contradictions
+5. Apply update:
+   - `proposal_required`: create `PersonUpdateProposal`
+   - `direct`: write only above confidence threshold and with no conflict
+6. Persist audit trail:
+   - record detector/classifier source and exact message provenance
+
+## Field-Specific Policy
+- `summary/profile`: generated from stable high-confidence aggregates only.
+- `revealed`: only explicit self-disclosures.
+- `likes/dislikes`: require explicit statement or repeated pattern.
+- `sentiment`: rolling value with recency decay; never absolute truth label.
+- `timezone`: explicit declaration preferred; behavioral inference secondary.
+- `last_interaction`: deterministic from most recent message timestamps.
+
+## Rollout
+1. Schema and settings models.
+2. Deterministic extractor pipeline and commands.
+3. Proposal queue + review flow.
+4. Optional Manticore semantic ranking layer.
+5. Backfill job for existing persons with safe rate limits.
+
+## Acceptance Criteria
+- Default enrichment path runs with zero OpenAI usage.
+- Person updates are traceable to concrete message evidence.
+- Config hierarchy behaves predictably (`user > group > global`).
+- Operators can switch between proposal and direct write modes per scope.
+
+## Out of Scope
+- Cross-user shared person graph.
+- Autonomous LLM-generated profile writing as default.
--- a/artifacts/plans/2026-03-07-settings-integrity-and-controls-reorg.md
+++ b/artifacts/plans/2026-03-07-settings-integrity-and-controls-reorg.md
@@ -0,0 +1,158 @@
+# Plan: Settings Integrity and Controls Reorganization
+
+## Objective
+Create a single coherent configuration model for Security, Commands, and Tasks so UI labels, enforcement behavior, docs, and navigation all match actual runtime behavior.
+
+## Current Integrity Findings
+
+### 1) Scope Registry and Enforcement Are Out of Sync
+- Gateway command routes enforce scopes that are not exposed in Fine-Grained Security Scopes:
+  - `gateway.contacts`
+  - `gateway.help`
+  - `gateway.whoami`
+- Fine-Grained Security Scopes currently expose only:
+  - `gateway.tasks`, `gateway.approval`, `tasks.submit`, `tasks.commands`, `command.bp`, `command.codex`, `command.claude`
+- Result: users cannot configure all enforced gateway capabilities from the UI.
+
+### 2) “Require Trusted Fingerprint” Semantics Are Incorrect
+- UI and labels imply trust-list based enforcement.
+- Runtime policy enforcement checks `UserXmppOmemoState.latest_client_key` equality, not `UserXmppOmemoTrustedKey` trust records.
+- Result: behavior is “match latest observed key,” not “require trusted fingerprint.”
+
+### 3) Command Surfaces Are Split Across Inconsistent Places
+- Command Routing UI create flow exposes command slugs: `bp`, `codex`.
+- Runtime command engine auto-bootstraps `claude` profile and bindings.
+- Security scopes include `command.claude`, but Command Routing create UI does not.
+- Result: commands are partially configurable depending on entrypoint.
+
+### 4) Task and Command Control Planes Interlock Implicitly, Not Explicitly
+- Task settings contain provider approval routing (Codex/Claude approver service/identifier).
+- Security permissions contain policy gates (`tasks.*`, `command.*`, `gateway.*`).
+- Command Routing controls profile/binding/variant policy.
+- These are tightly coupled but not represented as one layered model in UI/docs.
+
+### 5) Settings Shell Coverage Is Incomplete
+- Shared settings hierarchy nav is context-processor driven and route-name based.
+- Settings routes missing from modules/general/security group coverage include:
+  - `codex_settings`
+  - `codex_approval`
+  - `translation_preview`
+- Result: some settings pages can miss expected local tabs/title context.
+
+### 6) Documentation Drift
+- Undocumented or under-documented features now in production behavior:
+  - Fine-Grained Security Scopes + Global Scope Override
+  - OMEMO trust management and per-direction encryption toggles
+  - Business Plan Inbox under Settings Modules
+- Potentially misleading documentation:
+  - Security wording implies trusted-key enforcement that is not implemented.
+
+## Reorganization Principles
+1. One capability registry, reused by:
+   - Security Permissions UI
+   - command/task/gateway dispatch
+   - documentation generation
+2. One settings-shell contract for every `/settings/*` page:
+   - title
+   - category tabs
+   - breadcrumb
+3. Explicit layered model:
+   - Layer A: transport encryption/security
+   - Layer B: capability permissions (scope policy)
+   - Layer C: feature configuration (tasks/commands/providers)
+4. No hardcoded duplicated scope lists in multiple files.
+
+## Target Information Architecture
+
+### Security
+- `Encryption`: OMEMO transport controls + trust management.
+- `Permissions`: Fine-Grained Security Scopes (capability policy only).
+- `2FA`: account factor settings.
+
+### Modules
+- `Commands`: command profiles, bindings, variant policies.
+- `Tasks`: extraction/defaults/overrides/provider pipelines.
+- `Translation`: translation bridge settings.
+- `Availability`: adapter availability controls.
+- `Business Plans`: inbox/editor for generated artifacts.
+
+### General
+- `Notifications`
+- `System`
+- `Accessibility`
+
+### AI
+- `Models`
+- `Traces`
+
+## Phased Execution Plan
+
+## Phase 1: Canonical Capability Registry
+1. Add a central capability registry module (single source of truth):
+   - key
+   - label
+   - description
+   - group (`gateway`, `tasks`, `commands`, `agentic`, etc.)
+   - owning feature page URL
+2. Migrate SecurityPage scope rendering to this registry.
+3. Migrate gateway/command/task dispatchers to reference registry keys.
+4. Add automated integrity test:
+   - every enforced scope key must exist in registry
+   - every registry key marked user-configurable must appear in Permissions UI
+
+## Phase 2: Trusted Key Enforcement Correction
+1. Define authoritative trust policy behavior:
+   - `require_trusted_fingerprint` must validate against `UserXmppOmemoTrustedKey`.
+2. Preserve backwards compatibility via migration path:
+   - existing latest-key behavior can be temporarily represented as an optional fallback mode.
+3. Update labels/help to match exact behavior.
+4. Add tests:
+   - trusted key allows
+   - untrusted key denies
+   - unknown key denies
+
+## Phase 3: Commands/Tasks Control Plane Alignment
+1. Unify command surface definitions:
+   - Command Routing create/edit options include all supported command slugs (`bp`, `codex`, `claude`).
+2. Add explicit cross-links:
+   - Tasks settings references Command Routing and Permissions scopes directly.
+   - Command Routing references Permissions scopes affecting each profile.
+3. Introduce capability-impact preview panel:
+   - for each command/task action, show effective allow/deny by scope and channel.
+
+## Phase 4: Settings Shell Normalization
+1. Replace route-name allowlists in `settings_hierarchy_nav` with category mapping table.
+2. Ensure all `/settings/*` pages declare category + tab metadata.
+3. Include missing routes (`codex_settings`, `codex_approval`, `translation_preview`) in shell.
+4. Add test to fail when a `/settings/*` route lacks shell metadata.
+
+## Phase 5: Documentation Synchronization
+1. Add a settings matrix doc generated (or validated) from the capability registry:
+   - capability key
+   - UI location
+   - enforced by code path
+2. Update `README.md` and `INSTALL.md` security/modules sections.
+3. Add "policy semantics" section clarifying:
+   - encryption-required vs per-scope OMEMO requirements
+   - trusted key behavior
+   - global override precedence
+
+## Acceptance Criteria
+- Every enforced scope is user-visible/configurable (or intentionally internal and documented).
+- “Require Trusted Fingerprint” enforcement uses trust records, not only latest observed key.
+- Command Routing and runtime-supported command slugs are aligned.
+- All `/settings/*` pages show consistent settings shell navigation.
+- Security/tasks/commands docs reflect real behavior and pass integrity checks.
+
+## Risks and Mitigations
+- Risk: policy behavior change blocks existing workflows.
+  - Mitigation: add compatibility flag and staged rollout.
+- Risk: registry migration introduces missing scope mappings.
+  - Mitigation: integrity test that compares runtime-enforced keys vs registry.
+- Risk: UI complexity increase.
+  - Mitigation: keep layered model with concise, context-aware summaries.
+
+## Implementation Notes
+- Keep migration incremental; avoid big-bang rewrite.
+- Prioritize Phase 1 + Phase 2 first, because they are correctness and security semantics issues.
+- Do not add new transport-specific branches; keep service-agnostic evaluation path in policy engine.
--- a/auth_django.py
+++ b/auth_django.py
@@ -1,85 +0,0 @@
-# Create a debug log to confirm script execution
-
-import os
-import sys
-
-import django
-
-LOG_PATH = os.environ.get("AUTH_DEBUG_LOG", "/tmp/auth_debug.log")
-
-
-def log(data):
-    try:
-        with open(LOG_PATH, "a") as f:
-            f.write(f"{data}\n")
-    except Exception:
-        pass
-
-
-# Set up Django environment
-os.environ.setdefault("DJANGO_SETTINGS_MODULE", "app.settings")  # Adjust if needed
-django.setup()
-
-from django.contrib.auth import authenticate  # noqa: E402
-from django.contrib.auth.models import User  # noqa: E402
-
-
-def check_credentials(username, password):
-    """Authenticate user via Django"""
-    user = authenticate(username=username, password=password)
-    return user is not None and user.is_active
-
-
-def main():
-    """Process authentication requests from Prosody"""
-    while True:
-        try:
-            # Read a single line from stdin
-            line = sys.stdin.readline().strip()
-            if not line:
-                break  # Exit if input is empty (EOF)
-
-            # Log received command (for debugging)
-            # log(f"Received: {line}")
-
-            parts = line.split(":")
-            if len(parts) < 3:
-                log("Sending 0")
-                print("0", flush=True)  # Invalid format, return failure
-                continue
-
-            command, username, domain = parts[:3]
-            password = (
-                ":".join(parts[3:]) if len(parts) > 3 else None
-            )  # Reconstruct password
-
-            if command == "auth":
-                if password and check_credentials(username, password):
-                    log("Authentication success")
-                    log("Sent 1")
-                    print("1", flush=True)  # Success
-                else:
-                    log("Authentication failure")
-                    log("Sent 0")
-                    print("0", flush=True)  # Failure
-
-            elif command == "isuser":
-                if User.objects.filter(username=username).exists():
-                    print("1", flush=True)  # User exists
-                else:
-                    print("0", flush=True)  # User does not exist
-
-            elif command == "setpass":
-                print("0", flush=True)  # Not supported
-
-            else:
-                print("0", flush=True)  # Unknown command, return failure
-
-        except Exception as e:
-            # Log any unexpected errors
-            log(f"Error: {str(e)}\n")
-            print("0", flush=True)  # Return failure for any error
-
-
-if __name__ == "__main__":
-    main()
--- a/auth_django.sh
+++ b/auth_django.sh
@@ -1,10 +0,0 @@
-#!/bin/sh
-set -eu
-
-# Prosody external auth expects a single long-lived stdin/stdout process.
-# Keep one stable process chain and hand off with exec.
-exec podman exec -i gia sh -lc '
-  cd /code &&
-  . /venv/bin/activate &&
-  exec python -u auth_django.py
-'
--- a/core/assist/init.py
+++ b/core/assist/init.py
--- a/core/assist/engine.py
+++ b/core/assist/engine.py
@@ -0,0 +1,13 @@
+from __future__ import annotations
+
+from core.assist.repeat_answer import find_repeat_answer, learn_from_message
+from core.models import Message
+from core.tasks.engine import process_inbound_task_intelligence
+
+
+async def process_inbound_assist(message: Message) -> None:
+    if message is None:
+        return
+    await learn_from_message(message)
+    await find_repeat_answer(message.user, message)
+    await process_inbound_task_intelligence(message)
--- a/core/assist/repeat_answer.py
+++ b/core/assist/repeat_answer.py
@@ -0,0 +1,153 @@
+from __future__ import annotations
+
+import hashlib
+import re
+from dataclasses import dataclass
+
+from asgiref.sync import sync_to_async
+from django.utils import timezone
+
+from core.models import AnswerMemory, AnswerSuggestionEvent, Message
+
+_WORD_RE = re.compile(r"[^a-z0-9\s]+", re.IGNORECASE)
+
+
+@dataclass(slots=True)
+class RepeatAnswerSuggestion:
+    answer_memory_id: str
+    answer_text: str
+    score: float
+
+
+def _normalize_question(text: str) -> str:
+    body = str(text or "").strip().lower()
+    body = _WORD_RE.sub(" ", body)
+    body = re.sub(r"\s+", " ", body).strip()
+    return body
+
+
+def _fingerprint(text: str) -> str:
+    norm = _normalize_question(text)
+    if not norm:
+        return ""
+    return hashlib.sha1(norm.encode("utf-8")).hexdigest()
+
+
+def _is_question(text: str) -> bool:
+    body = str(text or "").strip()
+    if not body:
+        return False
+    low = body.lower()
+    return body.endswith("?") or low.startswith(
+        (
+            "what",
+            "why",
+            "how",
+            "when",
+            "where",
+            "who",
+            "can ",
+            "do ",
+            "did ",
+            "is ",
+            "are ",
+        )
+    )
+
+
+def _is_group_channel(message: Message) -> bool:
+    channel = str(getattr(message, "source_chat_id", "") or "").strip().lower()
+    if channel.endswith("@g.us"):
+        return True
+    return (
+        str(getattr(message, "source_service", "") or "").strip().lower() == "xmpp"
+        and "conference." in channel
+    )
+
+
+async def learn_from_message(message: Message) -> None:
+    if message is None:
+        return
+    text = str(message.text or "").strip()
+    if not text:
+        return
+    if dict(message.message_meta or {}).get("origin_tag"):
+        return
+
+    # Build memory by linking obvious reply answers to prior questions.
+    if message.reply_to_id and message.reply_to:
+        q_text = str(message.reply_to.text or "").strip()
+        if _is_question(q_text):
+            fp = _fingerprint(q_text)
+            if fp:
+                await sync_to_async(AnswerMemory.objects.create)(
+                    user=message.user,
+                    service=message.source_service or "web",
+                    channel_identifier=message.source_chat_id or "",
+                    question_fingerprint=fp,
+                    question_text=q_text,
+                    answer_message=message,
+                    answer_text=text,
+                    confidence_meta={"source": "reply_pair"},
+                )
+
+
+async def find_repeat_answer(user, message: Message) -> RepeatAnswerSuggestion | None:
+    if message is None:
+        return None
+    if not _is_group_channel(message):
+        return None
+    if dict(message.message_meta or {}).get("origin_tag"):
+        return None
+    text = str(message.text or "").strip()
+    if not _is_question(text):
+        return None
+
+    fp = _fingerprint(text)
+    if not fp:
+        return None
+
+    # channel cooldown for repeated suggestions in short windows
+    cooldown_cutoff = timezone.now() - timezone.timedelta(minutes=3)
+    cooldown_exists = await sync_to_async(
+        lambda: AnswerSuggestionEvent.objects.filter(
+            user=user,
+            message__source_service=message.source_service,
+            message__source_chat_id=message.source_chat_id,
+            status="suggested",
+            created_at__gte=cooldown_cutoff,
+        ).exists()
+    )()
+    if cooldown_exists:
+        return None
+
+    memory = await sync_to_async(
+        lambda: AnswerMemory.objects.filter(
+            user=user,
+            service=message.source_service or "web",
+            channel_identifier=message.source_chat_id or "",
+            question_fingerprint=fp,
+        )
+        .order_by("-created_at")
+        .first()
+    )()
+    if not memory:
+        return None
+
+    answer = str(memory.answer_text or "").strip()
+    if not answer:
+        return None
+
+    score = 0.99
+    await sync_to_async(AnswerSuggestionEvent.objects.create)(
+        user=user,
+        message=message,
+        status="suggested",
+        candidate_answer=memory,
+        score=score,
+    )
+    return RepeatAnswerSuggestion(
+        answer_memory_id=str(memory.id),
+        answer_text=answer,
+        score=score,
+    )
--- a/core/clients/init.py
+++ b/core/clients/init.py
@@ -12,8 +12,7 @@ class ClientBase(ABC):
        self.log.info(f"{self.service.capitalize()} client initialising...")

    @abstractmethod
-    def start(self):
-        ...
+    def start(self): ...

    # @abstractmethod
    # async def send_message(self, recipient, message):
--- a/core/clients/signal.py
+++ b/core/clients/signal.py
--- a/core/clients/signalapi.py
+++ b/core/clients/signalapi.py
@@ -1,5 +1,7 @@
 import asyncio
 import base64
+import logging
+import re

 import aiohttp
 import orjson
@@ -7,6 +9,43 @@ import requests
 from django.conf import settings
 from rest_framework import status

+log = logging.getLogger(__name__)
+SIGNAL_UUID_PATTERN = re.compile(
+    r"^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$",
+    re.IGNORECASE,
+)
+
+
+def _safe_parse_send_response(payload_value) -> int | bool:
+    payload = payload_value
+    if isinstance(payload_value, str):
+        try:
+            payload = orjson.loads(payload_value)
+        except orjson.JSONDecodeError:
+            return False
+    if not isinstance(payload, dict):
+        return False
+    try:
+        ts = payload.get("timestamp")
+        return int(ts) if ts else False
+    except (TypeError, ValueError):
+        return False
+
+
+def normalize_signal_recipient(recipient: str) -> str:
+    raw = str(recipient or "").strip()
+    if not raw:
+        return ""
+    if SIGNAL_UUID_PATTERN.fullmatch(raw):
+        return raw
+    if raw.startswith("+"):
+        digits = re.sub(r"[^0-9]", "", raw)
+        return f"+{digits}" if digits else raw
+    digits_only = re.sub(r"[^0-9]", "", raw)
+    if digits_only and raw.isdigit():
+        return f"+{digits_only}"
+    return raw
+

 async def start_typing(uuid):
    base = getattr(settings, "SIGNAL_HTTP_URL", "http://signal:8080").rstrip("/")
@@ -70,7 +109,9 @@ async def download_and_encode_base64(file_url, filename, content_type, session=N
        return None


-async def send_message_raw(recipient_uuid, text=None, attachments=None):
+async def send_message_raw(
+    recipient_uuid, text=None, attachments=None, metadata=None, detailed=False
+):
    """
    Sends a message using the Signal REST API, ensuring attachment links are not included in the text body.

@@ -85,11 +126,13 @@ async def send_message_raw(recipient_uuid, text=None, attachments=None):
    base = getattr(settings, "SIGNAL_HTTP_URL", "http://signal:8080").rstrip("/")
    url = f"{base}/v2/send"

+    normalized_recipient = normalize_signal_recipient(recipient_uuid)
    data = {
-        "recipients": [recipient_uuid],
+        "recipients": [normalized_recipient],
        "number": settings.SIGNAL_NUMBER,
        "base64_attachments": [],
    }
+    meta = dict(metadata or {})

    async def _attachment_to_base64(attachment, session):
        row = dict(attachment or {})
@@ -132,15 +175,76 @@ async def send_message_raw(recipient_uuid, text=None, attachments=None):
    if text:
        data["message"] = text

-    async with aiohttp.ClientSession() as session:
-        async with session.post(url, json=data) as response:
-            response_text = await response.text()
-            response_status = response.status
+    quote_timestamp = int(meta.get("quote_timestamp") or 0)
+    quote_author = str(meta.get("quote_author") or "").strip()
+    quote_text = str(meta.get("quote_text") or "").strip()
+    has_quote = quote_timestamp > 0 and bool(quote_author)

-            if response_status == status.HTTP_201_CREATED:
-                ts = orjson.loads(response_text).get("timestamp", None)
-                return ts if ts else False
-            return False
+    payloads = [dict(data)]
+    if has_quote:
+        flat_quote_payload = dict(data)
+        flat_quote_payload["quote_timestamp"] = int(quote_timestamp)
+        flat_quote_payload["quote_author"] = quote_author
+        if quote_text:
+            flat_quote_payload["quote_message"] = quote_text
+
+        nested_quote_payload = dict(data)
+        nested_quote_payload["quote"] = {
+            "id": int(quote_timestamp),
+            "author": quote_author,
+        }
+        if quote_text:
+            nested_quote_payload["quote"]["text"] = quote_text
+
+        payloads = [flat_quote_payload, nested_quote_payload, dict(data)]
+
+    async with aiohttp.ClientSession() as session:
+        for index, payload in enumerate(payloads):
+            async with session.post(url, json=payload) as response:
+                response_text = await response.text()
+                response_status = response.status
+                if response_status == status.HTTP_201_CREATED:
+                    ts = orjson.loads(response_text).get("timestamp", None)
+                    return ts if ts else False
+                if index == len(payloads) - 1:
+                    log.warning(
+                        "Signal send failed status=%s recipient=%s body=%s",
+                        response_status,
+                        normalized_recipient,
+                        response_text[:300],
+                    )
+                    if detailed:
+                        return {
+                            "ok": False,
+                            "status": int(response_status),
+                            "error": str(response_text or "").strip()[:500],
+                            "recipient": normalized_recipient,
+                        }
+                    return False
+                if response_status not in {
+                    status.HTTP_400_BAD_REQUEST,
+                    status.HTTP_422_UNPROCESSABLE_ENTITY,
+                }:
+                    log.warning(
+                        "Signal send failed early status=%s recipient=%s body=%s",
+                        response_status,
+                        normalized_recipient,
+                        response_text[:300],
+                    )
+                    if detailed:
+                        return {
+                            "ok": False,
+                            "status": int(response_status),
+                            "error": str(response_text or "").strip()[:500],
+                            "recipient": normalized_recipient,
+                        }
+                    return False
+                log.warning(
+                    "signal send quote payload rejected (%s), trying fallback shape: %s",
+                    response_status,
+                    response_text[:200],
+                )
+    return False


 async def send_reaction(
@@ -152,13 +256,17 @@ async def send_reaction(
 ):
    base = getattr(settings, "SIGNAL_HTTP_URL", "http://signal:8080").rstrip("/")
    sender_number = settings.SIGNAL_NUMBER
-    if not recipient_uuid or not target_timestamp:
+    normalized_recipient = normalize_signal_recipient(recipient_uuid)
+    normalized_target_author = normalize_signal_recipient(
+        str(target_author or normalized_recipient)
+    )
+    if not normalized_recipient or not target_timestamp:
        return False

    payload = {
-        "recipient": recipient_uuid,
+        "recipient": normalized_recipient,
        "reaction": str(emoji or ""),
-        "target_author": str(target_author or recipient_uuid),
+        "target_author": normalized_target_author,
        "timestamp": int(target_timestamp),
        "remove": bool(remove),
    }
@@ -269,7 +377,7 @@ def send_message_raw_sync(recipient_uuid, text=None, attachments=None):
    base = getattr(settings, "SIGNAL_HTTP_URL", "http://signal:8080").rstrip("/")
    url = f"{base}/v2/send"
    data = {
-        "recipients": [recipient_uuid],
+        "recipients": [normalize_signal_recipient(recipient_uuid)],
        "number": settings.SIGNAL_NUMBER,
        "base64_attachments": [],
    }
@@ -303,8 +411,8 @@ def send_message_raw_sync(recipient_uuid, text=None, attachments=None):
        response.status_code == status.HTTP_201_CREATED
    ):  # Signal server returns 201 on success
        try:
-            ts = orjson.loads(response.text).get("timestamp", None)
-            return ts if ts else False
-        except orjson.JSONDecodeError:
-            return False
+            payload = response.json()
+        except ValueError:
+            payload = {}
+        return _safe_parse_send_response(payload)
    return False  # If response status is not 201
--- a/core/clients/transport.py
+++ b/core/clients/transport.py
@@ -1,8 +1,11 @@
 import asyncio
 import base64
 import io
+import os
 import secrets
+import shutil
 import time
+from pathlib import Path
 from typing import Any
 from urllib.parse import quote_plus

@@ -15,6 +18,11 @@ from django.core.cache import cache

 from core.clients import signalapi
 from core.messaging import media_bridge
+from core.security.attachments import (
+    validate_attachment_metadata,
+    validate_attachment_url,
+)
+from core.transports.capabilities import supports, unsupported_reason
 from core.util import logs

 log = logs.get_logger("transport")
@@ -30,6 +38,10 @@ def _service_key(service: str) -> str:
    return str(service or "").strip().lower()


+def _capability_checks_enabled() -> bool:
+    return bool(getattr(settings, "CAPABILITY_ENFORCEMENT_ENABLED", True))
+
+
 def _runtime_key(service: str) -> str:
    return f"gia:service:runtime:{_service_key(service)}"

@@ -433,23 +445,6 @@ def list_accounts(service: str):
    Return account identifiers for service UI list.
    """
    service_key = _service_key(service)
-    if service_key == "signal":
-        import requests
-
-        base = str(getattr(settings, "SIGNAL_HTTP_URL", "http://signal:8080")).rstrip(
-            "/"
-        )
-        try:
-            response = requests.get(f"{base}/v1/accounts", timeout=20)
-            if not response.ok:
-                return []
-            payload = orjson.loads(response.text or "[]")
-            if isinstance(payload, list):
-                return payload
-        except Exception:
-            return []
-        return []
-
    state = get_runtime_state(service_key)
    accounts = state.get("accounts") or []
    if service_key == "whatsapp" and not accounts:
@@ -479,6 +474,57 @@ def _account_key(value: str) -> str:
    return raw


+def _wipe_signal_cli_local_state() -> bool:
+    """
+    Best-effort local signal-cli state reset for json-rpc deployments where
+    REST account delete endpoints are unavailable.
+    """
+    config_roots = []
+    base_dir = getattr(settings, "BASE_DIR", None)
+    if base_dir:
+        config_roots.append(str(Path(base_dir) / "signal-cli-config"))
+    config_roots.extend(
+        [
+            "/code/signal-cli-config",
+            "/signal-cli-config",
+            "/home/.local/share/signal-cli",
+        ]
+    )
+    removed_any = False
+    seen_roots = set()
+    for root in config_roots:
+        root = str(root or "").strip()
+        if not root or root in seen_roots:
+            continue
+        seen_roots.add(root)
+        if not os.path.isdir(root):
+            continue
+        try:
+            entries = os.listdir(root)
+        except Exception:
+            continue
+        for entry in entries:
+            if not entry:
+                continue
+            # Keep runtime configuration scaffold; wipe account/pairing state.
+            if entry in {"jsonrpc2.yml", "jsonrpc.yml"}:
+                continue
+            path = os.path.join(root, entry)
+            if os.path.isdir(path):
+                try:
+                    shutil.rmtree(path)
+                    removed_any = True
+                except Exception:
+                    continue
+            else:
+                try:
+                    os.remove(path)
+                    removed_any = True
+                except Exception:
+                    continue
+    return removed_any
+
+
 def unlink_account(service: str, account: str) -> bool:
    service_key = _service_key(service)
    account_value = str(account or "").strip()
@@ -492,14 +538,37 @@ def unlink_account(service: str, account: str) -> bool:
            "/"
        )
        target = quote_plus(account_value)
+        unlinked = False
        for path in (f"/v1/accounts/{target}", f"/v1/account/{target}"):
            try:
                response = requests.delete(f"{base}{path}", timeout=20)
                if response.ok:
-                    return True
+                    unlinked = True
+                    break
            except Exception:
                continue
-        return False
+        if unlinked:
+            return True
+        wiped = _wipe_signal_cli_local_state()
+        if not wiped:
+            return False
+        # Best-effort verification: if the REST API still reports the same account,
+        # the runtime likely still holds active linked state and the UI should not
+        # claim relink is ready yet.
+        remaining_accounts = list_accounts("signal")
+        for row in remaining_accounts:
+            if isinstance(row, dict):
+                candidate = (
+                    row.get("number")
+                    or row.get("id")
+                    or row.get("jid")
+                    or row.get("account")
+                )
+            else:
+                candidate = row
+            if _account_key(str(candidate or "")) == _account_key(account_value):
+                return False
+        return True

    if service_key in {"whatsapp", "instagram"}:
        state = get_runtime_state(service_key)
@@ -614,17 +683,21 @@ async def _normalize_gateway_attachment(service: str, row: dict, session):
    if isinstance(content, memoryview):
        content = content.tobytes()
    if isinstance(content, bytes):
+        filename, content_type = validate_attachment_metadata(
+            filename=normalized.get("filename") or "attachment.bin",
+            content_type=normalized.get("content_type") or "application/octet-stream",
+            size=normalized.get("size") or len(content),
+        )
        blob_key = media_bridge.put_blob(
            service=service,
            content=content,
-            filename=normalized.get("filename") or "attachment.bin",
-            content_type=normalized.get("content_type") or "application/octet-stream",
+            filename=filename,
+            content_type=content_type,
        )
        return {
            "blob_key": blob_key,
-            "filename": normalized.get("filename") or "attachment.bin",
-            "content_type": normalized.get("content_type")
-            or "application/octet-stream",
+            "filename": filename,
+            "content_type": content_type,
            "size": normalized.get("size") or len(content),
        }

@@ -634,33 +707,39 @@ async def _normalize_gateway_attachment(service: str, row: dict, session):
    source_url = normalized.get("url")
    if source_url:
        try:
-            async with session.get(source_url) as response:
+            safe_url = validate_attachment_url(source_url)
+            async with session.get(safe_url) as response:
                if response.status == 200:
                    payload = await response.read()
-                    blob_key = media_bridge.put_blob(
-                        service=service,
-                        content=payload,
+                    filename, content_type = validate_attachment_metadata(
                        filename=normalized.get("filename")
-                        or source_url.rstrip("/").split("/")[-1]
+                        or safe_url.rstrip("/").split("/")[-1]
                        or "attachment.bin",
                        content_type=normalized.get("content_type")
                        or response.headers.get(
                            "Content-Type", "application/octet-stream"
                        ),
+                        size=normalized.get("size") or len(payload),
+                    )
+                    blob_key = media_bridge.put_blob(
+                        service=service,
+                        content=payload,
+                        filename=filename,
+                        content_type=content_type,
                    )
                    return {
                        "blob_key": blob_key,
-                        "filename": normalized.get("filename")
-                        or source_url.rstrip("/").split("/")[-1]
-                        or "attachment.bin",
-                        "content_type": normalized.get("content_type")
-                        or response.headers.get(
-                            "Content-Type", "application/octet-stream"
-                        ),
+                        "filename": filename,
+                        "content_type": content_type,
                        "size": normalized.get("size") or len(payload),
                    }
-        except Exception:
-            log.warning("%s attachment fetch failed for %s", service, source_url)
+        except Exception as exc:
+            log.warning(
+                "%s attachment fetch failed for %s: %s",
+                service,
+                source_url,
+                exc,
+            )
    return normalized


@@ -711,12 +790,25 @@ async def send_message_raw(
    Unified outbound send path used by models/views/UR.
    """
    service_key = _service_key(service)
+    if _capability_checks_enabled() and not supports(service_key, "send"):
+        reason = unsupported_reason(service_key, "send")
+        log.warning(
+            "capability-check failed service=%s feature=send: %s",
+            service_key,
+            reason,
+        )
+        return False
    if service_key == "signal":
        prepared_attachments = await prepare_outbound_attachments(
            service_key, attachments or []
        )
-        result = await signalapi.send_message_raw(recipient, text, prepared_attachments)
        meta = dict(metadata or {})
+        result = await signalapi.send_message_raw(
+            recipient,
+            text,
+            prepared_attachments,
+            metadata=meta,
+        )
        xmpp_source_id = str(meta.get("xmpp_source_id") or "").strip()
        if xmpp_source_id and result:
            from core.models import PersonIdentifier
@@ -764,9 +856,7 @@ async def send_message_raw(
                runtime_result = await runtime_client.send_message_raw(
                    recipient,
                    text=text,
-                    attachments=await prepare_outbound_attachments(
-                        service_key, attachments or []
-                    ),
+                    attachments=attachments or [],
                    metadata=dict(metadata or {}),
                )
                if runtime_result is not False and runtime_result is not None:
@@ -775,11 +865,8 @@ async def send_message_raw(
                log.warning("%s runtime send failed: %s", service_key, exc)
        # Web/UI process cannot access UR in-process runtime client directly.
        # Hand off send to UR via shared cache command queue.
-        prepared_attachments = await prepare_outbound_attachments(
-            service_key, attachments or []
-        )
        command_attachments = []
-        for att in prepared_attachments:
+        for att in (attachments or []):
            row = dict(att or {})
            # Keep payload cache-friendly and avoid embedding raw bytes.
            for key in ("content",):
@@ -847,6 +934,14 @@ async def send_reaction(
    remove: bool = False,
 ):
    service_key = _service_key(service)
+    if _capability_checks_enabled() and not supports(service_key, "reactions"):
+        reason = unsupported_reason(service_key, "reactions")
+        log.warning(
+            "capability-check failed service=%s feature=reactions: %s",
+            service_key,
+            reason,
+        )
+        return False
    if not str(emoji or "").strip() and not remove:
        return False

@@ -917,6 +1012,13 @@ async def send_reaction(

 async def start_typing(service: str, recipient: str):
    service_key = _service_key(service)
+    if _capability_checks_enabled() and not supports(service_key, "typing"):
+        log.warning(
+            "capability-check failed service=%s feature=typing: %s",
+            service_key,
+            unsupported_reason(service_key, "typing"),
+        )
+        return False
    if service_key == "signal":
        await signalapi.start_typing(recipient)
        return True
@@ -947,6 +1049,13 @@ async def start_typing(service: str, recipient: str):

 async def stop_typing(service: str, recipient: str):
    service_key = _service_key(service)
+    if _capability_checks_enabled() and not supports(service_key, "typing"):
+        log.warning(
+            "capability-check failed service=%s feature=typing: %s",
+            service_key,
+            unsupported_reason(service_key, "typing"),
+        )
+        return False
    if service_key == "signal":
        await signalapi.stop_typing(recipient)
        return True
@@ -982,9 +1091,8 @@ async def fetch_attachment(service: str, attachment_ref: dict):
    service_key = _service_key(service)
    if service_key == "signal":
        attachment_id = attachment_ref.get("id") or attachment_ref.get("attachment_id")
-        if not attachment_id:
-            return None
-        return await signalapi.fetch_signal_attachment(attachment_id)
+        if attachment_id:
+            return await signalapi.fetch_signal_attachment(attachment_id)

    runtime_client = get_runtime_client(service_key)
    if runtime_client and hasattr(runtime_client, "fetch_attachment"):
@@ -1000,21 +1108,27 @@ async def fetch_attachment(service: str, attachment_ref: dict):
    if blob_key:
        return media_bridge.get_blob(blob_key)
    if direct_url:
+        safe_url = validate_attachment_url(direct_url)
        timeout = aiohttp.ClientTimeout(total=20)
        async with aiohttp.ClientSession(timeout=timeout) as session:
-            async with session.get(direct_url) as response:
+            async with session.get(safe_url) as response:
                if response.status != 200:
                    return None
                content = await response.read()
-                return {
-                    "content": content,
-                    "content_type": response.headers.get(
+                filename, content_type = validate_attachment_metadata(
+                    filename=attachment_ref.get("filename")
+                    or safe_url.rstrip("/").split("/")[-1]
+                    or "attachment.bin",
+                    content_type=response.headers.get(
                        "Content-Type",
                        attachment_ref.get("content_type", "application/octet-stream"),
                    ),
-                    "filename": attachment_ref.get("filename")
-                    or direct_url.rstrip("/").split("/")[-1]
-                    or "attachment.bin",
+                    size=len(content),
+                )
+                return {
+                    "content": content,
+                    "content_type": content_type,
+                    "filename": filename,
                    "size": len(content),
                }
    return None
@@ -1054,7 +1168,7 @@ def get_link_qr(service: str, device_name: str):
        response = requests.get(
            f"{base}/v1/qrcodelink",
            params={"device_name": device},
-            timeout=20,
+            timeout=5,
        )
        response.raise_for_status()
        return response.content
--- a/core/clients/whatsapp.py
+++ b/core/clients/whatsapp.py
--- a/core/clients/xmpp.py
+++ b/core/clients/xmpp.py
--- a/core/commands/init.py
+++ b/core/commands/init.py
--- a/core/commands/base.py
+++ b/core/commands/base.py
@@ -0,0 +1,29 @@
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any
+
+
+@dataclass(slots=True)
+class CommandContext:
+    service: str
+    channel_identifier: str
+    message_id: str
+    user_id: int
+    message_text: str
+    payload: dict[str, Any] = field(default_factory=dict)
+
+
+@dataclass(slots=True)
+class CommandResult:
+    ok: bool
+    status: str = "ok"
+    error: str = ""
+    payload: dict[str, Any] = field(default_factory=dict)
+
+
+class CommandHandler:
+    slug = ""
+
+    async def execute(self, ctx: CommandContext) -> CommandResult:
+        raise NotImplementedError
--- a/core/commands/delivery.py
+++ b/core/commands/delivery.py
@@ -0,0 +1,129 @@
+from __future__ import annotations
+
+import time
+
+from asgiref.sync import sync_to_async
+
+from core.clients import transport
+from core.models import ChatSession, Message
+
+STATUS_VISIBLE_SOURCE_SERVICES = {"web", "xmpp", "signal", "whatsapp"}
+
+
+def chunk_for_transport(text: str, limit: int = 3000) -> list[str]:
+    body = str(text or "").strip()
+    if not body:
+        return []
+    if len(body) <= limit:
+        return [body]
+    parts = []
+    remaining = body
+    while len(remaining) > limit:
+        cut = remaining.rfind("\n\n", 0, limit)
+        if cut < int(limit * 0.45):
+            cut = remaining.rfind("\n", 0, limit)
+        if cut < int(limit * 0.35):
+            cut = limit
+        parts.append(remaining[:cut].rstrip())
+        remaining = remaining[cut:].lstrip()
+    if remaining:
+        parts.append(remaining)
+    return [part for part in parts if part]
+
+
+async def post_status_in_source(
+    trigger_message: Message, text: str, origin_tag: str
+) -> bool:
+    service = str(trigger_message.source_service or "").strip().lower()
+    if service not in STATUS_VISIBLE_SOURCE_SERVICES:
+        return False
+    if service == "web":
+        await sync_to_async(Message.objects.create)(
+            user=trigger_message.user,
+            session=trigger_message.session,
+            sender_uuid="",
+            text=text,
+            ts=int(time.time() * 1000),
+            custom_author="BOT",
+            source_service="web",
+            source_chat_id=trigger_message.source_chat_id or "",
+            message_meta={"origin_tag": origin_tag},
+        )
+        return True
+    # For non-web, route through transport raw API.
+    if not str(trigger_message.source_chat_id or "").strip():
+        return False
+    try:
+        await transport.send_message_raw(
+            service,
+            str(trigger_message.source_chat_id or "").strip(),
+            text=text,
+            attachments=[],
+            metadata={"origin_tag": origin_tag},
+        )
+        return True
+    except Exception:
+        return False
+
+
+async def post_to_channel_binding(
+    trigger_message: Message,
+    binding_service: str,
+    binding_channel_identifier: str,
+    text: str,
+    origin_tag: str,
+    command_slug: str,
+) -> bool:
+    service = str(binding_service or "").strip().lower()
+    channel_identifier = str(binding_channel_identifier or "").strip()
+    if service == "web":
+        session = None
+        if (
+            channel_identifier
+            and channel_identifier == str(trigger_message.source_chat_id or "").strip()
+        ):
+            session = trigger_message.session
+        if session is None and channel_identifier:
+            session = await sync_to_async(
+                lambda: ChatSession.objects.filter(
+                    user=trigger_message.user,
+                    identifier__identifier=channel_identifier,
+                )
+                .order_by("-last_interaction")
+                .first()
+            )()
+        if session is None:
+            session = trigger_message.session
+        await sync_to_async(Message.objects.create)(
+            user=trigger_message.user,
+            session=session,
+            sender_uuid="",
+            text=text,
+            ts=int(time.time() * 1000),
+            custom_author="BOT",
+            source_service="web",
+            source_chat_id=channel_identifier
+            or str(trigger_message.source_chat_id or ""),
+            message_meta={"origin_tag": origin_tag},
+        )
+        return True
+    try:
+        chunks = chunk_for_transport(text, limit=3000)
+        if not chunks:
+            return False
+        for chunk in chunks:
+            ts = await transport.send_message_raw(
+                service,
+                channel_identifier,
+                text=chunk,
+                attachments=[],
+                metadata={
+                    "origin_tag": origin_tag,
+                    "command_slug": command_slug,
+                },
+            )
+            if not ts:
+                return False
+        return True
+    except Exception:
+        return False
--- a/core/commands/engine.py
+++ b/core/commands/engine.py
@@ -0,0 +1,501 @@
+from __future__ import annotations
+
+from asgiref.sync import sync_to_async
+
+from core.commands.base import CommandContext, CommandResult
+from core.commands.handlers.bp import (
+    BPCommandHandler,
+    bp_reply_is_optional_for_trigger,
+    bp_subcommands_enabled,
+    bp_trigger_matches,
+)
+from core.commands.handlers.claude import ClaudeCommandHandler, claude_trigger_matches
+from core.commands.handlers.codex import CodexCommandHandler, codex_trigger_matches
+from core.commands.policies import ensure_variant_policies_for_profile
+from core.commands.registry import get as get_handler
+from core.commands.registry import register
+from core.messaging.reply_sync import is_mirrored_origin
+from core.models import (
+    CommandAction,
+    CommandChannelBinding,
+    CommandProfile,
+    Message,
+    PersonIdentifier,
+)
+from core.security.command_policy import CommandSecurityContext, evaluate_command_policy
+from core.tasks.chat_defaults import ensure_default_source_for_chat
+from core.util import logs
+
+log = logs.get_logger("command_engine")
+
+_REGISTERED = False
+
+
+def _channel_variants(service: str, channel_identifier: str) -> list[str]:
+    value = str(channel_identifier or "").strip()
+    if not value:
+        return []
+    variants = [value]
+    svc = str(service or "").strip().lower()
+    if svc == "whatsapp":
+        bare = value.split("@", 1)[0].strip()
+        if bare and bare not in variants:
+            variants.append(bare)
+        group = f"{bare}@g.us" if bare else ""
+        if group and group not in variants:
+            variants.append(group)
+    return variants
+
+
+def _canonical_channel_identifier(service: str, channel_identifier: str) -> str:
+    value = str(channel_identifier or "").strip()
+    if not value:
+        return ""
+    if str(service or "").strip().lower() == "whatsapp":
+        return value.split("@", 1)[0].strip()
+    return value
+
+
+def _signal_identifier_rank(identifier_value: str) -> int:
+    identifier_text = str(identifier_value or "").strip()
+    if not identifier_text:
+        return 99
+    if identifier_text.startswith("group."):
+        return 0
+    if identifier_text.startswith("+"):
+        return 1
+    return 2
+
+
+def _expand_service_channel_variants(
+    user_id: int,
+    service: str,
+    identifiers: list[str],
+) -> list[str]:
+    variants: list[str] = []
+    for identifier in identifiers:
+        for value in _channel_variants(service, identifier):
+            if value and value not in variants:
+                variants.append(value)
+    if str(service or "").strip().lower() != "signal" or not variants:
+        return variants
+    person_ids = list(
+        PersonIdentifier.objects.filter(
+            user_id=user_id,
+            service="signal",
+            identifier__in=variants,
+        )
+        .values_list("person_id", flat=True)
+        .distinct()
+    )
+    if not person_ids:
+        return variants
+    alias_rows = list(
+        PersonIdentifier.objects.filter(
+            user_id=user_id,
+            service="signal",
+            person_id__in=person_ids,
+        ).values_list("identifier", flat=True)
+    )
+    for value in alias_rows:
+        cleaned = str(value or "").strip()
+        if cleaned and cleaned not in variants:
+            variants.append(cleaned)
+    variants.sort(key=lambda value: (_signal_identifier_rank(value), value))
+    return variants
+
+
+def _preferred_channel_identifier(service: str, identifiers: list[str]) -> str:
+    cleaned = [str(value or "").strip() for value in identifiers if str(value or "").strip()]
+    if not cleaned:
+        return ""
+    if str(service or "").strip().lower() == "signal":
+        cleaned.sort(key=lambda value: (_signal_identifier_rank(value), value))
+        return cleaned[0]
+    return cleaned[0]
+
+
+def _effective_bootstrap_scope(
+    ctx: CommandContext,
+    trigger_message: Message,
+) -> tuple[str, str]:
+    service = str(ctx.service or "").strip().lower()
+    identifier = str(ctx.channel_identifier or "").strip()
+    if service != "web":
+        return service, identifier
+    session_identifier = getattr(
+        getattr(trigger_message, "session", None), "identifier", None
+    )
+    fallback_service = (
+        str(getattr(session_identifier, "service", "") or "").strip().lower()
+    )
+    fallback_identifier = str(
+        getattr(session_identifier, "identifier", "") or ""
+    ).strip()
+    if fallback_service and fallback_identifier and fallback_service != "web":
+        return fallback_service, fallback_identifier
+    return service, identifier
+
+
+def _ensure_bp_profile(user_id: int) -> CommandProfile:
+    profile, _ = CommandProfile.objects.get_or_create(
+        user_id=user_id,
+        slug="bp",
+        defaults={
+            "name": "Business Plan",
+            "enabled": True,
+            "trigger_token": ".bp",
+            "reply_required": True,
+            "exact_match_only": True,
+            "window_scope": "conversation",
+            "visibility_mode": "status_in_source",
+        },
+    )
+    updated = False
+    if not profile.enabled:
+        profile.enabled = True
+        updated = True
+    if updated:
+        profile.save(update_fields=["enabled", "updated_at"])
+    if str(profile.trigger_token or "").strip() != ".bp":
+        profile.trigger_token = ".bp"
+        profile.save(update_fields=["trigger_token", "updated_at"])
+    for action_type, position in (
+        ("extract_bp", 0),
+        ("save_document", 1),
+        ("post_result", 2),
+    ):
+        action, created = CommandAction.objects.get_or_create(
+            profile=profile,
+            action_type=action_type,
+            defaults={"enabled": True, "position": position},
+        )
+        if (not created) and (not action.enabled):
+            action.enabled = True
+            action.save(update_fields=["enabled", "updated_at"])
+    ensure_variant_policies_for_profile(profile)
+    return profile
+
+
+def _ensure_codex_profile(user_id: int) -> CommandProfile:
+    profile, _ = CommandProfile.objects.get_or_create(
+        user_id=user_id,
+        slug="codex",
+        defaults={
+            "name": "Codex",
+            "enabled": True,
+            "trigger_token": ".codex",
+            "reply_required": False,
+            "exact_match_only": False,
+            "window_scope": "conversation",
+            "visibility_mode": "status_in_source",
+        },
+    )
+    if not profile.enabled:
+        profile.enabled = True
+        profile.save(update_fields=["enabled", "updated_at"])
+    if str(profile.trigger_token or "").strip() != ".codex":
+        profile.trigger_token = ".codex"
+        profile.save(update_fields=["trigger_token", "updated_at"])
+    return profile
+
+
+def _ensure_claude_profile(user_id: int) -> CommandProfile:
+    profile, _ = CommandProfile.objects.get_or_create(
+        user_id=user_id,
+        slug="claude",
+        defaults={
+            "name": "Claude",
+            "enabled": True,
+            "trigger_token": ".claude",
+            "reply_required": False,
+            "exact_match_only": False,
+            "window_scope": "conversation",
+            "visibility_mode": "status_in_source",
+        },
+    )
+    if not profile.enabled:
+        profile.enabled = True
+        profile.save(update_fields=["enabled", "updated_at"])
+    if str(profile.trigger_token or "").strip() != ".claude":
+        profile.trigger_token = ".claude"
+        profile.save(update_fields=["trigger_token", "updated_at"])
+    return profile
+
+
+def _ensure_profile_for_slug(user_id: int, slug: str) -> CommandProfile | None:
+    if slug == "bp":
+        return _ensure_bp_profile(user_id)
+    if slug == "codex":
+        return _ensure_codex_profile(user_id)
+    if slug == "claude":
+        return _ensure_claude_profile(user_id)
+    return None
+
+
+def _detected_bootstrap_slugs(message_text: str) -> list[str]:
+    slugs: list[str] = []
+    if bp_trigger_matches(message_text, ".bp", False):
+        slugs.append("bp")
+    if codex_trigger_matches(message_text, ".codex", False):
+        slugs.append("codex")
+    if claude_trigger_matches(message_text, ".claude", False):
+        slugs.append("claude")
+    return slugs
+
+
+def _auto_setup_profile_bindings_for_first_command(
+    ctx: CommandContext,
+    trigger_message: Message,
+) -> None:
+    author = str(getattr(trigger_message, "custom_author", "") or "").strip().upper()
+    if author != "USER":
+        return
+    slugs = _detected_bootstrap_slugs(ctx.message_text)
+    if not slugs:
+        return
+    service, identifier = _effective_bootstrap_scope(ctx, trigger_message)
+    service = str(service or "").strip().lower()
+    canonical = _canonical_channel_identifier(service, identifier)
+    variants = _expand_service_channel_variants(ctx.user_id, service, [canonical])
+    canonical = _preferred_channel_identifier(service, variants) or canonical
+    if not service or not variants:
+        return
+    for slug in slugs:
+        profile = _ensure_profile_for_slug(ctx.user_id, slug)
+        if profile is None:
+            continue
+        already_enabled = CommandChannelBinding.objects.filter(
+            profile=profile,
+            enabled=True,
+            direction="ingress",
+            service=service,
+            channel_identifier__in=variants,
+        ).exists()
+        if already_enabled:
+            continue
+        for direction in ("ingress", "egress"):
+            binding, _ = CommandChannelBinding.objects.get_or_create(
+                profile=profile,
+                direction=direction,
+                service=service,
+                channel_identifier=canonical,
+                defaults={"enabled": True},
+            )
+            if not binding.enabled:
+                binding.enabled = True
+                binding.save(update_fields=["enabled", "updated_at"])
+            alternate_variants = [value for value in variants if value != canonical]
+            if alternate_variants:
+                CommandChannelBinding.objects.filter(
+                    profile=profile,
+                    direction=direction,
+                    service=service,
+                    channel_identifier__in=alternate_variants,
+                ).update(enabled=False)
+    ensure_default_source_for_chat(
+        user=trigger_message.user,
+        service=service,
+        channel_identifier=canonical,
+        message=trigger_message,
+    )
+
+
+def ensure_handlers_registered():
+    global _REGISTERED
+    if _REGISTERED:
+        return
+    register(BPCommandHandler())
+    register(CodexCommandHandler())
+    register(ClaudeCommandHandler())
+    _REGISTERED = True
+
+
+async def _eligible_profiles(ctx: CommandContext) -> list[CommandProfile]:
+    def _load():
+        trigger = (
+            Message.objects.select_related("session", "session__identifier")
+            .filter(id=ctx.message_id, user_id=ctx.user_id)
+            .first()
+        )
+        direct_variants = _expand_service_channel_variants(
+            ctx.user_id,
+            ctx.service,
+            [ctx.channel_identifier],
+        )
+        source_channel = str(getattr(trigger, "source_chat_id", "") or "").strip()
+        for expanded in _expand_service_channel_variants(
+            ctx.user_id,
+            ctx.service,
+            [source_channel],
+        ):
+            if expanded and expanded not in direct_variants:
+                direct_variants.append(expanded)
+        if not direct_variants:
+            return []
+        direct = list(
+            CommandProfile.objects.filter(
+                user_id=ctx.user_id,
+                enabled=True,
+                channel_bindings__enabled=True,
+                channel_bindings__direction="ingress",
+                channel_bindings__service=ctx.service,
+                channel_bindings__channel_identifier__in=direct_variants,
+            ).distinct()
+        )
+        if direct:
+            return direct
+        # Compose-originated messages use `web` service even when the
+        # underlying conversation is mapped to a platform identifier.
+        if str(ctx.service or "").strip().lower() != "web":
+            return []
+        identifier = getattr(getattr(trigger, "session", None), "identifier", None)
+        fallback_service = str(getattr(identifier, "service", "") or "").strip().lower()
+        fallback_identifier = str(getattr(identifier, "identifier", "") or "").strip()
+        fallback_variants = _expand_service_channel_variants(
+            ctx.user_id,
+            fallback_service,
+            [fallback_identifier],
+        )
+        for expanded in _expand_service_channel_variants(
+            ctx.user_id,
+            fallback_service,
+            [source_channel],
+        ):
+            if expanded and expanded not in fallback_variants:
+                fallback_variants.append(expanded)
+        if not fallback_service or not fallback_variants:
+            return []
+        return list(
+            CommandProfile.objects.filter(
+                user_id=ctx.user_id,
+                enabled=True,
+                channel_bindings__enabled=True,
+                channel_bindings__direction="ingress",
+                channel_bindings__service=fallback_service,
+                channel_bindings__channel_identifier__in=fallback_variants,
+            ).distinct()
+        )
+
+    return await sync_to_async(_load)()
+
+
+def _matches_trigger(profile: CommandProfile, text: str) -> bool:
+    if profile.slug == "bp" and bp_subcommands_enabled():
+        return bp_trigger_matches(
+            message_text=text,
+            trigger_token=profile.trigger_token,
+            exact_match_only=profile.exact_match_only,
+        )
+    if profile.slug == "codex":
+        return codex_trigger_matches(
+            message_text=text,
+            trigger_token=profile.trigger_token,
+            exact_match_only=profile.exact_match_only,
+        )
+    if profile.slug == "claude":
+        return claude_trigger_matches(
+            message_text=text,
+            trigger_token=profile.trigger_token,
+            exact_match_only=profile.exact_match_only,
+        )
+    body = str(text or "").strip()
+    trigger = str(profile.trigger_token or "").strip()
+    if not trigger:
+        return False
+    if profile.exact_match_only:
+        return body == trigger
+    return trigger in body
+
+
+async def process_inbound_message(ctx: CommandContext) -> list[CommandResult]:
+    ensure_handlers_registered()
+    trigger_message = await sync_to_async(
+        lambda: Message.objects.select_related("user", "session", "session__identifier")
+        .filter(id=ctx.message_id)
+        .first()
+    )()
+    if trigger_message is None:
+        return []
+    if is_mirrored_origin(trigger_message.message_meta):
+        return []
+    effective_service, effective_channel = _effective_bootstrap_scope(
+        ctx, trigger_message
+    )
+    security_context = CommandSecurityContext(
+        service=effective_service,
+        channel_identifier=effective_channel,
+        message_meta=dict(getattr(trigger_message, "message_meta", {}) or {}),
+        payload=dict(ctx.payload or {}),
+    )
+    await sync_to_async(_auto_setup_profile_bindings_for_first_command)(
+        ctx,
+        trigger_message,
+    )
+
+    profiles = await _eligible_profiles(ctx)
+    results: list[CommandResult] = []
+    for profile in profiles:
+        if not _matches_trigger(profile, ctx.message_text):
+            continue
+        decision = await sync_to_async(evaluate_command_policy)(
+            user=trigger_message.user,
+            scope_key=f"command.{profile.slug}",
+            context=security_context,
+        )
+        if not decision.allowed:
+            results.append(
+                CommandResult(
+                    ok=False,
+                    status="skipped",
+                    error=f"policy_denied:{decision.code}",
+                    payload={
+                        "profile": profile.slug,
+                        "scope": f"command.{profile.slug}",
+                        "reason": decision.reason,
+                    },
+                )
+            )
+            continue
+        if profile.reply_required and trigger_message.reply_to_id is None:
+            if (
+                profile.slug == "bp"
+                and bp_subcommands_enabled()
+                and bp_reply_is_optional_for_trigger(ctx.message_text)
+            ):
+                pass
+            else:
+                results.append(
+                    CommandResult(
+                        ok=False,
+                        status="skipped",
+                        error="reply_required",
+                        payload={"profile": profile.slug},
+                    )
+                )
+                continue
+        handler = get_handler(profile.slug)
+        if handler is None:
+            results.append(
+                CommandResult(
+                    ok=False,
+                    status="failed",
+                    error=f"missing_handler:{profile.slug}",
+                )
+            )
+            continue
+        try:
+            result = await handler.execute(ctx)
+            results.append(result)
+        except Exception as exc:
+            log.exception(
+                "command execution failed for profile=%s: %s", profile.slug, exc
+            )
+            results.append(
+                CommandResult(
+                    ok=False,
+                    status="failed",
+                    error=f"handler_exception:{exc}",
+                )
+            )
+    return results
--- a/core/commands/handlers/init.py
+++ b/core/commands/handlers/init.py
--- a/core/commands/handlers/bp.py
+++ b/core/commands/handlers/bp.py
@@ -0,0 +1,715 @@
+from __future__ import annotations
+
+import re
+import time
+
+from asgiref.sync import sync_to_async
+from django.conf import settings
+
+from core.commands.base import CommandContext, CommandHandler, CommandResult
+from core.commands.delivery import post_status_in_source, post_to_channel_binding
+from core.commands.policies import BP_VARIANT_META, load_variant_policy
+from core.messaging import ai as ai_runner
+from core.messaging.text_export import plain_text_blob
+from core.messaging.utils import messages_to_string
+from core.models import (
+    AI,
+    BusinessPlanDocument,
+    BusinessPlanRevision,
+    CommandAction,
+    CommandChannelBinding,
+    CommandRun,
+    CommandVariantPolicy,
+    Message,
+)
+
+_BP_ROOT_RE = re.compile(r"^\s*(?:\.bp\b|#bp#?)\s*$", re.IGNORECASE)
+_BP_SET_RE = re.compile(
+    r"^\s*(?:\.bp\s+set\b|#bp\s+set#?)(?P<rest>.*)$",
+    re.IGNORECASE | re.DOTALL,
+)
+_BP_SET_RANGE_RE = re.compile(
+    r"^\s*(?:\.bp\s+set\s+range\b|#bp\s+set\s+range#?)(?:.*)$",
+    re.IGNORECASE | re.DOTALL,
+)
+
+
+class BPParsedCommand(dict):
+    @property
+    def command(self) -> str | None:
+        value = self.get("command")
+        return str(value) if value else None
+
+    @property
+    def remainder_text(self) -> str:
+        return str(self.get("remainder_text") or "")
+
+
+def parse_bp_subcommand(text: str) -> BPParsedCommand:
+    body = str(text or "")
+    if _BP_SET_RANGE_RE.match(body):
+        return BPParsedCommand(command="set_range", remainder_text="")
+    match = _BP_SET_RE.match(body)
+    if match:
+        return BPParsedCommand(
+            command="set", remainder_text=str(match.group("rest") or "").strip()
+        )
+    return BPParsedCommand(command=None, remainder_text="")
+
+
+def bp_subcommands_enabled() -> bool:
+    raw = getattr(settings, "BP_SUBCOMMANDS_V1", True)
+    if raw is None:
+        return True
+    return bool(raw)
+
+
+def bp_trigger_matches(
+    message_text: str, trigger_token: str, exact_match_only: bool
+) -> bool:
+    body = str(message_text or "").strip()
+    trigger = str(trigger_token or "").strip()
+    parsed = parse_bp_subcommand(body)
+    if parsed.command and bp_subcommands_enabled():
+        return True
+    if _BP_ROOT_RE.match(body):
+        return True
+    if not trigger:
+        return False
+    if exact_match_only:
+        return body.lower() == trigger.lower()
+    return trigger.lower() in body.lower()
+
+
+def bp_reply_is_optional_for_trigger(message_text: str) -> bool:
+    parsed = parse_bp_subcommand(message_text)
+    return parsed.command == "set"
+
+
+def _bp_system_prompt():
+    return (
+        "Create a structured business plan using the given template. "
+        "Follow the template section order exactly. "
+        "If data is missing, write concise assumptions and risks. "
+        "Return markdown only."
+    )
+
+
+def _clamp_transcript(transcript: str, max_chars: int) -> str:
+    text = str(transcript or "")
+    if max_chars <= 0 or len(text) <= max_chars:
+        return text
+    head_size = min(2000, max_chars // 3)
+    tail_size = max(0, max_chars - head_size - 140)
+    omitted = len(text) - head_size - tail_size
+    return (
+        text[:head_size].rstrip()
+        + f"\n\n[... truncated {max(0, omitted)} chars ...]\n\n"
+        + text[-tail_size:].lstrip()
+    )
+
+
+class BPCommandHandler(CommandHandler):
+    slug = "bp"
+
+    def _variant_key_for_text(self, text: str) -> str:
+        parsed = parse_bp_subcommand(text)
+        if parsed.command == "set":
+            return "bp_set"
+        if parsed.command == "set_range":
+            return "bp_set_range"
+        return "bp"
+
+    def _variant_display_name(self, variant_key: str) -> str:
+        meta = BP_VARIANT_META.get(str(variant_key or "").strip(), {})
+        return str(meta.get("name") or variant_key or "bp")
+
+    async def _effective_policy(
+        self,
+        *,
+        profile,
+        variant_key: str,
+        action_types: set[str],
+    ) -> dict:
+        policy = await sync_to_async(load_variant_policy)(profile, variant_key)
+        if isinstance(policy, CommandVariantPolicy):
+            return {
+                "enabled": bool(policy.enabled),
+                "generation_mode": str(policy.generation_mode or "verbatim"),
+                "send_plan_to_egress": bool(policy.send_plan_to_egress)
+                and ("post_result" in action_types),
+                "send_status_to_source": bool(policy.send_status_to_source)
+                or str(profile.visibility_mode or "") == "status_in_source",
+                "send_status_to_egress": bool(policy.send_status_to_egress),
+                "store_document": bool(getattr(policy, "store_document", True)),
+            }
+        return {
+            "enabled": True,
+            "generation_mode": "ai" if variant_key == "bp" else "verbatim",
+            "send_plan_to_egress": "post_result" in action_types,
+            "send_status_to_source": str(profile.visibility_mode or "")
+            == "status_in_source",
+            "send_status_to_egress": False,
+            "store_document": True,
+        }
+
+    async def _fanout(self, run: CommandRun, text: str) -> dict:
+        profile = run.profile
+        trigger = await sync_to_async(
+            lambda: Message.objects.select_related("session", "user")
+            .filter(id=run.trigger_message_id)
+            .first()
+        )()
+        if trigger is None:
+            return {"sent_bindings": 0, "failed_bindings": 0}
+        bindings = await sync_to_async(list)(
+            CommandChannelBinding.objects.filter(
+                profile=profile,
+                enabled=True,
+                direction="egress",
+            )
+        )
+        sent_bindings = 0
+        failed_bindings = 0
+        for binding in bindings:
+            ok = await post_to_channel_binding(
+                trigger_message=trigger,
+                binding_service=binding.service,
+                binding_channel_identifier=binding.channel_identifier,
+                text=text,
+                origin_tag=f"bp:{run.id}",
+                command_slug=self.slug,
+            )
+            if ok:
+                sent_bindings += 1
+            else:
+                failed_bindings += 1
+        return {"sent_bindings": sent_bindings, "failed_bindings": failed_bindings}
+
+    async def _fanout_status(self, run: CommandRun, text: str) -> dict:
+        profile = run.profile
+        trigger = await sync_to_async(
+            lambda: Message.objects.select_related("session", "user")
+            .filter(id=run.trigger_message_id)
+            .first()
+        )()
+        if trigger is None:
+            return {"sent_bindings": 0, "failed_bindings": 0}
+        bindings = await sync_to_async(list)(
+            CommandChannelBinding.objects.filter(
+                profile=profile,
+                enabled=True,
+                direction="egress",
+            )
+        )
+        sent_bindings = 0
+        failed_bindings = 0
+        for binding in bindings:
+            ok = await post_to_channel_binding(
+                trigger_message=trigger,
+                binding_service=binding.service,
+                binding_channel_identifier=binding.channel_identifier,
+                text=text,
+                origin_tag=f"bp-status-egress:{run.id}",
+                command_slug=self.slug,
+            )
+            if ok:
+                sent_bindings += 1
+            else:
+                failed_bindings += 1
+        return {"sent_bindings": sent_bindings, "failed_bindings": failed_bindings}
+
+    async def _load_window(self, trigger: Message, anchor: Message) -> list[Message]:
+        return await sync_to_async(list)(
+            Message.objects.filter(
+                user=trigger.user,
+                session=trigger.session,
+                ts__gte=int(anchor.ts or 0),
+                ts__lte=int(trigger.ts or 0),
+            )
+            .order_by("ts")
+            .select_related(
+                "session", "session__identifier", "session__identifier__person"
+            )
+        )
+
+    def _annotation(
+        self, mode: str, message_count: int, has_addendum: bool = False
+    ) -> str:
+        if mode == "set" and has_addendum:
+            return "Generated from 1 message + 1 addendum."
+        if message_count == 1:
+            return "Generated from 1 message."
+        return f"Generated from {int(message_count)} messages."
+
+    async def _persist_document(
+        self,
+        *,
+        run: CommandRun,
+        trigger: Message,
+        profile,
+        anchor: Message | None,
+        content: str,
+        mode: str,
+        source_message_ids: list[str],
+        annotation: str,
+    ) -> BusinessPlanDocument:
+        payload = {
+            "mode": mode,
+            "source_message_ids": list(source_message_ids),
+            "annotation": annotation,
+        }
+        document = await sync_to_async(BusinessPlanDocument.objects.create)(
+            user=trigger.user,
+            command_profile=profile,
+            source_service=trigger.source_service or "web",
+            source_channel_identifier=trigger.source_chat_id or "",
+            trigger_message=trigger,
+            anchor_message=anchor,
+            title=f"Business Plan {time.strftime('%Y-%m-%d %H:%M:%S')}",
+            status="draft",
+            content_markdown=content,
+            structured_payload=payload,
+        )
+        await sync_to_async(BusinessPlanRevision.objects.create)(
+            document=document,
+            editor_user=trigger.user,
+            content_markdown=content,
+            structured_payload=payload,
+        )
+        run.result_ref = document
+        await sync_to_async(run.save)(update_fields=["result_ref", "updated_at"])
+        return document
+
+    async def _execute_set_or_range(
+        self,
+        *,
+        trigger: Message,
+        run: CommandRun,
+        profile,
+        policy: dict,
+        variant_key: str,
+        parsed: BPParsedCommand,
+    ) -> CommandResult:
+        mode = str(parsed.command or "")
+        remainder = parsed.remainder_text
+        anchor = trigger.reply_to
+
+        if mode == "set_range":
+            if anchor is None:
+                run.status = "failed"
+                run.error = "bp_set_range_requires_reply_target"
+                await sync_to_async(run.save)(
+                    update_fields=["status", "error", "updated_at"]
+                )
+                return CommandResult(ok=False, status="failed", error=run.error)
+            rows = await self._load_window(trigger, anchor)
+            deterministic_content = plain_text_blob(rows)
+            if not deterministic_content.strip():
+                run.status = "failed"
+                run.error = "bp_set_range_empty_content"
+                await sync_to_async(run.save)(
+                    update_fields=["status", "error", "updated_at"]
+                )
+                return CommandResult(ok=False, status="failed", error=run.error)
+            if str(policy.get("generation_mode") or "verbatim") == "ai":
+                ai_obj = await sync_to_async(
+                    lambda: AI.objects.filter(user=trigger.user).first()
+                )()
+                if ai_obj is None:
+                    run.status = "failed"
+                    run.error = "ai_not_configured"
+                    await sync_to_async(run.save)(
+                        update_fields=["status", "error", "updated_at"]
+                    )
+                    return CommandResult(ok=False, status="failed", error=run.error)
+                prompt = [
+                    {
+                        "role": "system",
+                        "content": (
+                            "Transform source chat text into a structured business plan in markdown. "
+                            "Do not reference any user template."
+                        ),
+                    },
+                    {"role": "user", "content": deterministic_content},
+                ]
+                try:
+                    content = str(
+                        await ai_runner.run_prompt(
+                            prompt,
+                            ai_obj,
+                            operation="command_bp_set_range_extract",
+                        )
+                        or ""
+                    ).strip()
+                except Exception as exc:
+                    run.status = "failed"
+                    run.error = f"bp_ai_failed:{exc}"
+                    await sync_to_async(run.save)(
+                        update_fields=["status", "error", "updated_at"]
+                    )
+                    return CommandResult(ok=False, status="failed", error=run.error)
+                if not content:
+                    run.status = "failed"
+                    run.error = "empty_ai_response"
+                    await sync_to_async(run.save)(
+                        update_fields=["status", "error", "updated_at"]
+                    )
+                    return CommandResult(ok=False, status="failed", error=run.error)
+            else:
+                content = deterministic_content
+            annotation = self._annotation("set_range", len(rows))
+            doc = None
+            if bool(policy.get("store_document", True)):
+                doc = await self._persist_document(
+                    run=run,
+                    trigger=trigger,
+                    profile=profile,
+                    anchor=anchor,
+                    content=content,
+                    mode="set_range",
+                    source_message_ids=[str(row.id) for row in rows],
+                    annotation=annotation,
+                )
+        elif mode == "set":
+            source_ids: list[str] = []
+            if anchor is not None and not remainder:
+                content = str(anchor.text or "").strip() or "(no text)"
+                source_ids.append(str(anchor.id))
+                has_addendum = False
+            elif anchor is not None and remainder:
+                base = str(anchor.text or "").strip() or "(no text)"
+                content = (
+                    f"{base}\n" "--- Addendum (newer message text) ---\n" f"{remainder}"
+                )
+                source_ids.extend([str(anchor.id), str(trigger.id)])
+                has_addendum = True
+            elif remainder:
+                content = remainder
+                source_ids.append(str(trigger.id))
+                has_addendum = False
+            else:
+                run.status = "failed"
+                run.error = "bp_set_empty_content"
+                await sync_to_async(run.save)(
+                    update_fields=["status", "error", "updated_at"]
+                )
+                return CommandResult(ok=False, status="failed", error=run.error)
+
+            if str(policy.get("generation_mode") or "verbatim") == "ai":
+                ai_obj = await sync_to_async(
+                    lambda: AI.objects.filter(user=trigger.user).first()
+                )()
+                if ai_obj is None:
+                    run.status = "failed"
+                    run.error = "ai_not_configured"
+                    await sync_to_async(run.save)(
+                        update_fields=["status", "error", "updated_at"]
+                    )
+                    return CommandResult(ok=False, status="failed", error=run.error)
+                prompt = [
+                    {
+                        "role": "system",
+                        "content": (
+                            "Transform source chat text into a structured business plan in markdown. "
+                            "Do not reference any user template."
+                        ),
+                    },
+                    {"role": "user", "content": content},
+                ]
+                try:
+                    ai_content = str(
+                        await ai_runner.run_prompt(
+                            prompt,
+                            ai_obj,
+                            operation="command_bp_set_extract",
+                        )
+                        or ""
+                    ).strip()
+                except Exception as exc:
+                    run.status = "failed"
+                    run.error = f"bp_ai_failed:{exc}"
+                    await sync_to_async(run.save)(
+                        update_fields=["status", "error", "updated_at"]
+                    )
+                    return CommandResult(ok=False, status="failed", error=run.error)
+                if not ai_content:
+                    run.status = "failed"
+                    run.error = "empty_ai_response"
+                    await sync_to_async(run.save)(
+                        update_fields=["status", "error", "updated_at"]
+                    )
+                    return CommandResult(ok=False, status="failed", error=run.error)
+                content = ai_content
+
+            annotation = self._annotation(
+                "set", 1 if not has_addendum else 2, has_addendum
+            )
+            doc = None
+            if bool(policy.get("store_document", True)):
+                doc = await self._persist_document(
+                    run=run,
+                    trigger=trigger,
+                    profile=profile,
+                    anchor=anchor,
+                    content=content,
+                    mode="set",
+                    source_message_ids=source_ids,
+                    annotation=annotation,
+                )
+        else:
+            run.status = "failed"
+            run.error = "bp_unknown_subcommand"
+            await sync_to_async(run.save)(
+                update_fields=["status", "error", "updated_at"]
+            )
+            return CommandResult(ok=False, status="failed", error=run.error)
+
+        fanout_stats = {"sent_bindings": 0, "failed_bindings": 0}
+        if bool(policy.get("send_plan_to_egress")):
+            fanout_body = f"{content}\n\n{annotation}".strip()
+            fanout_stats = await self._fanout(run, fanout_body)
+
+        sent_count = int(fanout_stats.get("sent_bindings") or 0)
+        failed_count = int(fanout_stats.get("failed_bindings") or 0)
+        status_text = (
+            f"[bp:{self._variant_display_name(variant_key)}:{policy.get('generation_mode')}] "
+            f"{annotation.strip()} "
+            f"{'Saved as ' + doc.title + ' · ' if doc else 'Not saved (store_document disabled) · '}"
+            f"fanout sent:{sent_count}"
+        ).strip()
+        if failed_count:
+            status_text += f" failed:{failed_count}"
+
+        if bool(policy.get("send_status_to_source")):
+            await post_status_in_source(
+                trigger_message=trigger,
+                text=status_text,
+                origin_tag=f"bp-status:{trigger.id}",
+            )
+        if bool(policy.get("send_status_to_egress")):
+            await self._fanout_status(run, status_text)
+
+        run.status = "ok"
+        run.error = ""
+        await sync_to_async(run.save)(update_fields=["status", "error", "updated_at"])
+        return CommandResult(
+            ok=True,
+            status="ok",
+            payload={"document_id": str(doc.id) if doc else ""},
+        )
+
+    async def _execute_legacy_ai(
+        self,
+        *,
+        trigger: Message,
+        run: CommandRun,
+        profile,
+        policy: dict,
+        variant_key: str,
+    ) -> CommandResult:
+        if trigger.reply_to_id is None:
+            run.status = "failed"
+            run.error = "bp_requires_reply_target"
+            await sync_to_async(run.save)(
+                update_fields=["status", "error", "updated_at"]
+            )
+            return CommandResult(ok=False, status="failed", error=run.error)
+
+        anchor = trigger.reply_to
+        rows = await self._load_window(trigger, anchor)
+        transcript = messages_to_string(
+            rows,
+            author_rewrites={"USER": "Operator", "BOT": "Assistant"},
+        )
+        max_transcript_chars = int(
+            getattr(settings, "BP_MAX_TRANSCRIPT_CHARS", 12000) or 12000
+        )
+        transcript = _clamp_transcript(transcript, max_transcript_chars)
+        default_template = (
+            "Business Plan:\n"
+            "- Objective\n"
+            "- Audience\n"
+            "- Offer\n"
+            "- GTM\n"
+            "- Risks"
+        )
+        template_text = profile.template_text or default_template
+        max_template_chars = int(
+            getattr(settings, "BP_MAX_TEMPLATE_CHARS", 5000) or 5000
+        )
+        template_text = str(template_text or "")[:max_template_chars]
+        generation_mode = str(policy.get("generation_mode") or "ai")
+        if generation_mode == "verbatim":
+            summary = plain_text_blob(rows)
+            if not summary.strip():
+                run.status = "failed"
+                run.error = "bp_verbatim_empty_content"
+                await sync_to_async(run.save)(
+                    update_fields=["status", "error", "updated_at"]
+                )
+                return CommandResult(ok=False, status="failed", error=run.error)
+        else:
+            ai_obj = await sync_to_async(
+                lambda: AI.objects.filter(user=trigger.user).first()
+            )()
+            if ai_obj is None:
+                run.status = "failed"
+                run.error = "ai_not_configured"
+                await sync_to_async(run.save)(
+                    update_fields=["status", "error", "updated_at"]
+                )
+                return CommandResult(ok=False, status="failed", error=run.error)
+
+            prompt = [
+                {"role": "system", "content": _bp_system_prompt()},
+                {
+                    "role": "user",
+                    "content": (
+                        "Template:\n"
+                        f"{template_text}\n\n"
+                        "Messages:\n"
+                        f"{transcript}"
+                    ),
+                },
+            ]
+            try:
+                summary = str(
+                    await ai_runner.run_prompt(
+                        prompt, ai_obj, operation="command_bp_extract"
+                    )
+                    or ""
+                ).strip()
+                if not summary:
+                    raise RuntimeError("empty_ai_response")
+            except Exception as exc:
+                run.status = "failed"
+                run.error = f"bp_ai_failed:{exc}"
+                await sync_to_async(run.save)(
+                    update_fields=["status", "error", "updated_at"]
+                )
+                return CommandResult(ok=False, status="failed", error=run.error)
+
+        annotation = self._annotation("legacy", len(rows))
+        document = None
+        if bool(policy.get("store_document", True)):
+            document = await self._persist_document(
+                run=run,
+                trigger=trigger,
+                profile=profile,
+                anchor=anchor,
+                content=summary,
+                mode="legacy_ai",
+                source_message_ids=[str(row.id) for row in rows],
+                annotation=annotation,
+            )
+
+        fanout_stats = {"sent_bindings": 0, "failed_bindings": 0}
+        if bool(policy.get("send_plan_to_egress")):
+            fanout_stats = await self._fanout(run, summary)
+
+        sent_count = int(fanout_stats.get("sent_bindings") or 0)
+        failed_count = int(fanout_stats.get("failed_bindings") or 0)
+        status_text = (
+            f"[bp:{self._variant_display_name(variant_key)}:{generation_mode}] "
+            f"Generated business plan: "
+            f"{document.title if document else 'not saved (store_document disabled)'} "
+            f"· fanout sent:{sent_count}"
+        )
+        if failed_count:
+            status_text += f" failed:{failed_count}"
+
+        if bool(policy.get("send_status_to_source")):
+            await post_status_in_source(
+                trigger_message=trigger,
+                text=status_text,
+                origin_tag=f"bp-status:{trigger.id}",
+            )
+        if bool(policy.get("send_status_to_egress")):
+            await self._fanout_status(run, status_text)
+
+        run.status = "ok"
+        run.error = ""
+        await sync_to_async(run.save)(update_fields=["status", "error", "updated_at"])
+        return CommandResult(
+            ok=True,
+            status="ok",
+            payload={"document_id": str(document.id) if document else ""},
+        )
+
+    async def execute(self, ctx: CommandContext) -> CommandResult:
+        trigger = await sync_to_async(
+            lambda: Message.objects.select_related("user", "session")
+            .filter(id=ctx.message_id)
+            .first()
+        )()
+        if trigger is None:
+            return CommandResult(ok=False, status="failed", error="trigger_not_found")
+
+        profile = await sync_to_async(
+            lambda: trigger.user.commandprofile_set.filter(
+                slug=self.slug, enabled=True
+            ).first()
+        )()
+        if profile is None:
+            return CommandResult(ok=False, status="skipped", error="profile_missing")
+
+        actions = await sync_to_async(list)(
+            CommandAction.objects.filter(profile=profile, enabled=True).order_by(
+                "position", "id"
+            )
+        )
+        action_types = {row.action_type for row in actions}
+        if "extract_bp" not in action_types:
+            return CommandResult(
+                ok=False, status="skipped", error="extract_bp_disabled"
+            )
+
+        run, created = await sync_to_async(CommandRun.objects.get_or_create)(
+            profile=profile,
+            trigger_message=trigger,
+            defaults={"user": trigger.user, "status": "running"},
+        )
+        if not created and run.status in {"ok", "running"}:
+            return CommandResult(
+                ok=True,
+                status="ok",
+                payload={"document_id": str(run.result_ref_id or "")},
+            )
+
+        run.status = "running"
+        run.error = ""
+        await sync_to_async(run.save)(update_fields=["status", "error", "updated_at"])
+
+        variant_key = self._variant_key_for_text(ctx.message_text)
+        policy = await self._effective_policy(
+            profile=profile,
+            variant_key=variant_key,
+            action_types=action_types,
+        )
+        if not bool(policy.get("enabled")):
+            run.status = "skipped"
+            run.error = f"variant_disabled:{variant_key}"
+            await sync_to_async(run.save)(
+                update_fields=["status", "error", "updated_at"]
+            )
+            return CommandResult(ok=False, status="skipped", error=run.error)
+
+        parsed = parse_bp_subcommand(ctx.message_text)
+        if parsed.command and bp_subcommands_enabled():
+            return await self._execute_set_or_range(
+                trigger=trigger,
+                run=run,
+                profile=profile,
+                policy=policy,
+                variant_key=variant_key,
+                parsed=parsed,
+            )
+
+        return await self._execute_legacy_ai(
+            trigger=trigger,
+            run=run,
+            profile=profile,
+            policy=policy,
+            variant_key=variant_key,
+        )
--- a/core/commands/handlers/claude.py
+++ b/core/commands/handlers/claude.py
@@ -0,0 +1,629 @@
+from __future__ import annotations
+
+import hashlib
+import re
+
+from asgiref.sync import sync_to_async
+from django.utils import timezone
+
+from core.commands.base import CommandContext, CommandHandler, CommandResult
+from core.commands.delivery import post_status_in_source
+from core.messaging.text_export import plain_text_blob
+from core.models import (
+    ChatTaskSource,
+    CodexPermissionRequest,
+    CodexRun,
+    CommandProfile,
+    DerivedTask,
+    ExternalSyncEvent,
+    Message,
+    TaskProject,
+    TaskProviderConfig,
+)
+from core.tasks.codex_approval import queue_codex_event_with_pre_approval
+from core.tasks.codex_support import channel_variants, resolve_external_chat_id
+
+_CLAUDE_DEFAULT_RE = re.compile(
+    r"^\s*(?:\.claude\b|#claude#?)(?P<body>.*)$",
+    re.IGNORECASE | re.DOTALL,
+)
+_CLAUDE_PLAN_RE = re.compile(
+    r"^\s*(?:\.claude\s+plan\b|#claude\s+plan#?)(?P<body>.*)$",
+    re.IGNORECASE | re.DOTALL,
+)
+_CLAUDE_STATUS_RE = re.compile(
+    r"^\s*(?:\.claude\s+status\b|#claude\s+status#?)\s*$", re.IGNORECASE
+)
+_CLAUDE_APPROVE_DENY_RE = re.compile(
+    r"^\s*(?:\.claude|#claude)\s+(?P<action>approve|deny)\s+(?P<approval_key>[A-Za-z0-9._:-]+)#?\s*$",
+    re.IGNORECASE,
+)
+_PROJECT_TOKEN_RE = re.compile(r"\[\s*project\s*:\s*([^\]]+)\]", re.IGNORECASE)
+_REFERENCE_RE = re.compile(r"(?<!\w)#([A-Za-z0-9_-]+)\b")
+
+
+class ClaudeParsedCommand(dict):
+    @property
+    def command(self) -> str | None:
+        value = self.get("command")
+        return str(value) if value else None
+
+    @property
+    def body_text(self) -> str:
+        return str(self.get("body_text") or "")
+
+    @property
+    def approval_key(self) -> str:
+        return str(self.get("approval_key") or "")
+
+
+def parse_claude_command(text: str) -> ClaudeParsedCommand:
+    body = str(text or "")
+    m = _CLAUDE_APPROVE_DENY_RE.match(body)
+    if m:
+        return ClaudeParsedCommand(
+            command=str(m.group("action") or "").strip().lower(),
+            body_text="",
+            approval_key=str(m.group("approval_key") or "").strip(),
+        )
+    if _CLAUDE_STATUS_RE.match(body):
+        return ClaudeParsedCommand(command="status", body_text="", approval_key="")
+    m = _CLAUDE_PLAN_RE.match(body)
+    if m:
+        return ClaudeParsedCommand(
+            command="plan",
+            body_text=str(m.group("body") or "").strip(),
+            approval_key="",
+        )
+    m = _CLAUDE_DEFAULT_RE.match(body)
+    if m:
+        return ClaudeParsedCommand(
+            command="default",
+            body_text=str(m.group("body") or "").strip(),
+            approval_key="",
+        )
+    return ClaudeParsedCommand(command=None, body_text="", approval_key="")
+
+
+def claude_trigger_matches(
+    message_text: str, trigger_token: str, exact_match_only: bool
+) -> bool:
+    body = str(message_text or "").strip()
+    parsed = parse_claude_command(body)
+    if parsed.command:
+        return True
+    trigger = str(trigger_token or "").strip()
+    if not trigger:
+        return False
+    if exact_match_only:
+        return body.lower() == trigger.lower()
+    return trigger.lower() in body.lower()
+
+
+class ClaudeCommandHandler(CommandHandler):
+    slug = "claude"
+    _provider_name = "claude_cli"
+    _approval_prefix = "claude_approval"
+
+    async def _load_trigger(self, message_id: str) -> Message | None:
+        return await sync_to_async(
+            lambda: Message.objects.select_related(
+                "user", "session", "session__identifier", "reply_to"
+            )
+            .filter(id=message_id)
+            .first()
+        )()
+
+    def _effective_scope(self, trigger: Message) -> tuple[str, str]:
+        service = str(getattr(trigger, "source_service", "") or "").strip().lower()
+        channel = str(getattr(trigger, "source_chat_id", "") or "").strip()
+        identifier = getattr(getattr(trigger, "session", None), "identifier", None)
+        fallback_service = str(getattr(identifier, "service", "") or "").strip().lower()
+        fallback_identifier = str(getattr(identifier, "identifier", "") or "").strip()
+        if (
+            service == "web"
+            and fallback_service
+            and fallback_identifier
+            and fallback_service != "web"
+        ):
+            return fallback_service, fallback_identifier
+        return service or "web", channel
+
+    async def _mapped_sources(
+        self, user, service: str, channel: str
+    ) -> list[ChatTaskSource]:
+        variants = channel_variants(service, channel)
+        if not variants:
+            return []
+        return await sync_to_async(list)(
+            ChatTaskSource.objects.filter(
+                user=user,
+                enabled=True,
+                service=service,
+                channel_identifier__in=variants,
+            ).select_related("project", "epic")
+        )
+
+    async def _linked_task_from_reply(
+        self, user, reply_to: Message | None
+    ) -> DerivedTask | None:
+        if reply_to is None:
+            return None
+        by_origin = await sync_to_async(
+            lambda: DerivedTask.objects.filter(user=user, origin_message=reply_to)
+            .select_related("project", "epic")
+            .order_by("-created_at")
+            .first()
+        )()
+        if by_origin is not None:
+            return by_origin
+        return await sync_to_async(
+            lambda: DerivedTask.objects.filter(
+                user=user, events__source_message=reply_to
+            )
+            .select_related("project", "epic")
+            .order_by("-created_at")
+            .first()
+        )()
+
+    def _extract_project_token(self, body_text: str) -> tuple[str, str]:
+        text = str(body_text or "")
+        m = _PROJECT_TOKEN_RE.search(text)
+        if not m:
+            return "", text
+        token = str(m.group(1) or "").strip()
+        cleaned = _PROJECT_TOKEN_RE.sub("", text).strip()
+        return token, cleaned
+
+    def _extract_reference(self, body_text: str) -> str:
+        m = _REFERENCE_RE.search(str(body_text or ""))
+        if not m:
+            return ""
+        return str(m.group(1) or "").strip()
+
+    async def _resolve_task(
+        self, user, reference_code: str, reply_task: DerivedTask | None
+    ) -> DerivedTask | None:
+        if reference_code:
+            return await sync_to_async(
+                lambda: DerivedTask.objects.filter(
+                    user=user, reference_code=reference_code
+                )
+                .select_related("project", "epic")
+                .order_by("-created_at")
+                .first()
+            )()
+        return reply_task
+
+    async def _resolve_project(
+        self,
+        *,
+        user,
+        service: str,
+        channel: str,
+        task: DerivedTask | None,
+        reply_task: DerivedTask | None,
+        project_token: str,
+    ) -> tuple[TaskProject | None, str]:
+        if task is not None:
+            return task.project, ""
+        if reply_task is not None:
+            return reply_task.project, ""
+        if project_token:
+            project = await sync_to_async(
+                lambda: TaskProject.objects.filter(
+                    user=user, name__iexact=project_token
+                ).first()
+            )()
+            if project is not None:
+                return project, ""
+            return None, f"project_not_found:{project_token}"
+
+        mapped = await self._mapped_sources(user, service, channel)
+        project_ids = sorted({str(row.project_id) for row in mapped if row.project_id})
+        if len(project_ids) == 1:
+            project = next(
+                (
+                    row.project
+                    for row in mapped
+                    if str(row.project_id) == project_ids[0]
+                ),
+                None,
+            )
+            return project, ""
+        if len(project_ids) > 1:
+            return None, "project_required:[project:Name]"
+        return None, "project_unresolved"
+
+    async def _post_source_status(
+        self, trigger: Message, text: str, suffix: str
+    ) -> None:
+        await post_status_in_source(
+            trigger_message=trigger,
+            text=text,
+            origin_tag=f"claude-status:{suffix}",
+        )
+
+    async def _run_status(
+        self, trigger: Message, service: str, channel: str, project: TaskProject | None
+    ) -> CommandResult:
+        def _load_runs():
+            qs = CodexRun.objects.filter(user=trigger.user)
+            if service:
+                qs = qs.filter(source_service=service)
+            if channel:
+                qs = qs.filter(source_channel=channel)
+            if project is not None:
+                qs = qs.filter(project=project)
+            return list(qs.order_by("-created_at")[:10])
+
+        runs = await sync_to_async(_load_runs)()
+        if not runs:
+            await self._post_source_status(
+                trigger, "[claude] no recent runs for this scope.", "empty"
+            )
+            return CommandResult(ok=True, status="ok", payload={"count": 0})
+        lines = ["[claude] recent runs:"]
+        for row in runs:
+            ref = str(getattr(getattr(row, "task", None), "reference_code", "") or "-")
+            summary = str((row.result_payload or {}).get("summary") or "").strip()
+            summary_part = f" · {summary}" if summary else ""
+            lines.append(f"- {row.status} run={row.id} task=#{ref}{summary_part}")
+        await self._post_source_status(trigger, "\n".join(lines), "runs")
+        return CommandResult(ok=True, status="ok", payload={"count": len(runs)})
+
+    async def _run_approval_action(
+        self,
+        trigger: Message,
+        parsed: ClaudeParsedCommand,
+        current_service: str,
+        current_channel: str,
+    ) -> CommandResult:
+        cfg = await sync_to_async(
+            lambda: TaskProviderConfig.objects.filter(
+                user=trigger.user, provider=self._provider_name
+            ).first()
+        )()
+        settings_payload = dict(getattr(cfg, "settings", {}) or {})
+        approver_service = (
+            str(settings_payload.get("approver_service") or "").strip().lower()
+        )
+        approver_identifier = str(
+            settings_payload.get("approver_identifier") or ""
+        ).strip()
+        if not approver_service or not approver_identifier:
+            return CommandResult(
+                ok=False, status="failed", error="approver_channel_not_configured"
+            )
+
+        if str(current_service or "").strip().lower() != approver_service or str(
+            current_channel or ""
+        ).strip() not in set(channel_variants(approver_service, approver_identifier)):
+            return CommandResult(
+                ok=False,
+                status="failed",
+                error="approval_command_not_allowed_in_this_channel",
+            )
+
+        approval_key = parsed.approval_key
+        request = await sync_to_async(
+            lambda: CodexPermissionRequest.objects.select_related(
+                "codex_run", "external_sync_event"
+            )
+            .filter(user=trigger.user, approval_key=approval_key)
+            .first()
+        )()
+        if request is None:
+            return CommandResult(
+                ok=False, status="failed", error="approval_key_not_found"
+            )
+
+        now = timezone.now()
+        if parsed.command == "approve":
+            request.status = "approved"
+            request.resolved_at = now
+            request.resolved_by_identifier = current_channel
+            request.resolution_note = "approved via claude command"
+            await sync_to_async(request.save)(
+                update_fields=[
+                    "status",
+                    "resolved_at",
+                    "resolved_by_identifier",
+                    "resolution_note",
+                ]
+            )
+            if request.external_sync_event_id:
+                await sync_to_async(
+                    ExternalSyncEvent.objects.filter(
+                        id=request.external_sync_event_id
+                    ).update
+                )(
+                    status="ok",
+                    error="",
+                )
+            run = request.codex_run
+            run.status = "approved_waiting_resume"
+            run.error = ""
+            await sync_to_async(run.save)(
+                update_fields=["status", "error", "updated_at"]
+            )
+            source_service = str(run.source_service or "")
+            source_channel = str(run.source_channel or "")
+            resume_payload = dict(request.resume_payload or {})
+            resume_action = str(resume_payload.get("action") or "").strip().lower()
+            resume_provider_payload = dict(resume_payload.get("provider_payload") or {})
+            if resume_action and resume_provider_payload:
+                provider_payload = dict(resume_provider_payload)
+                provider_payload["codex_run_id"] = str(run.id)
+                provider_payload["source_service"] = source_service
+                provider_payload["source_channel"] = source_channel
+                event_action = resume_action
+                resume_idempotency_key = str(
+                    resume_payload.get("idempotency_key") or ""
+                ).strip()
+                resume_event_key = (
+                    resume_idempotency_key
+                    if resume_idempotency_key
+                    else f"{self._approval_prefix}:{approval_key}:approved"
+                )
+            else:
+                provider_payload = dict(
+                    run.request_payload.get("provider_payload") or {}
+                )
+                provider_payload.update(
+                    {
+                        "mode": "approval_response",
+                        "approval_key": approval_key,
+                        "resume_payload": dict(request.resume_payload or {}),
+                        "codex_run_id": str(run.id),
+                        "source_service": source_service,
+                        "source_channel": source_channel,
+                    }
+                )
+                event_action = "append_update"
+                resume_event_key = f"{self._approval_prefix}:{approval_key}:approved"
+            await sync_to_async(ExternalSyncEvent.objects.update_or_create)(
+                idempotency_key=resume_event_key,
+                defaults={
+                    "user": trigger.user,
+                    "task_id": run.task_id,
+                    "task_event_id": run.derived_task_event_id,
+                    "provider": self._provider_name,
+                    "status": "pending",
+                    "payload": {
+                        "action": event_action,
+                        "provider_payload": provider_payload,
+                    },
+                    "error": "",
+                },
+            )
+            return CommandResult(
+                ok=True,
+                status="ok",
+                payload={"approval_key": approval_key, "resolution": "approved"},
+            )
+
+        request.status = "denied"
+        request.resolved_at = now
+        request.resolved_by_identifier = current_channel
+        request.resolution_note = "denied via claude command"
+        await sync_to_async(request.save)(
+            update_fields=[
+                "status",
+                "resolved_at",
+                "resolved_by_identifier",
+                "resolution_note",
+            ]
+        )
+        if request.external_sync_event_id:
+            await sync_to_async(
+                ExternalSyncEvent.objects.filter(
+                    id=request.external_sync_event_id
+                ).update
+            )(
+                status="failed",
+                error="approval_denied",
+            )
+        run = request.codex_run
+        run.status = "denied"
+        run.error = "approval_denied"
+        await sync_to_async(run.save)(update_fields=["status", "error", "updated_at"])
+        await sync_to_async(ExternalSyncEvent.objects.update_or_create)(
+            idempotency_key=f"{self._approval_prefix}:{approval_key}:denied",
+            defaults={
+                "user": trigger.user,
+                "task_id": run.task_id,
+                "task_event_id": run.derived_task_event_id,
+                "provider": self._provider_name,
+                "status": "failed",
+                "payload": {
+                    "action": "append_update",
+                    "provider_payload": {
+                        "mode": "approval_response",
+                        "approval_key": approval_key,
+                        "codex_run_id": str(run.id),
+                    },
+                },
+                "error": "approval_denied",
+            },
+        )
+        return CommandResult(
+            ok=True,
+            status="ok",
+            payload={"approval_key": approval_key, "resolution": "denied"},
+        )
+
+    async def _create_submission(
+        self,
+        *,
+        trigger: Message,
+        mode: str,
+        body_text: str,
+        task: DerivedTask,
+        project: TaskProject,
+    ) -> CommandResult:
+        cfg = await sync_to_async(
+            lambda: TaskProviderConfig.objects.filter(
+                user=trigger.user, provider=self._provider_name, enabled=True
+            ).first()
+        )()
+        if cfg is None:
+            return CommandResult(
+                ok=False, status="failed", error="provider_disabled_or_missing"
+            )
+
+        service, channel = self._effective_scope(trigger)
+        external_chat_id = await sync_to_async(resolve_external_chat_id)(
+            user=trigger.user,
+            provider=self._provider_name,
+            service=service,
+            channel=channel,
+        )
+        payload = {
+            "task_id": str(task.id),
+            "reference_code": str(task.reference_code or ""),
+            "title": str(task.title or ""),
+            "external_key": str(task.external_key or ""),
+            "project_name": str(getattr(project, "name", "") or ""),
+            "epic_name": str(getattr(getattr(task, "epic", None), "name", "") or ""),
+            "source_service": service,
+            "source_channel": channel,
+            "external_chat_id": external_chat_id,
+            "origin_message_id": str(getattr(task, "origin_message_id", "") or ""),
+            "trigger_message_id": str(trigger.id),
+            "mode": mode,
+            "command_text": str(body_text or ""),
+        }
+        if mode == "plan":
+            anchor = trigger.reply_to
+            if anchor is None:
+                return CommandResult(
+                    ok=False, status="failed", error="reply_required_for_claude_plan"
+                )
+            rows = await sync_to_async(list)(
+                Message.objects.filter(
+                    user=trigger.user,
+                    session=trigger.session,
+                    ts__gte=int(anchor.ts or 0),
+                    ts__lte=int(trigger.ts or 0),
+                )
+                .order_by("ts")
+                .select_related(
+                    "session", "session__identifier", "session__identifier__person"
+                )
+            )
+            payload["reply_context"] = {
+                "anchor_message_id": str(anchor.id),
+                "trigger_message_id": str(trigger.id),
+                "message_ids": [str(row.id) for row in rows],
+                "content": plain_text_blob(rows),
+            }
+
+        run = await sync_to_async(CodexRun.objects.create)(
+            user=trigger.user,
+            task=task,
+            source_message=trigger,
+            project=project,
+            epic=getattr(task, "epic", None),
+            source_service=service,
+            source_channel=channel,
+            external_chat_id=external_chat_id,
+            status="waiting_approval",
+            request_payload={
+                "action": "append_update",
+                "provider_payload": dict(payload),
+            },
+            result_payload={},
+            error="",
+        )
+        payload["codex_run_id"] = str(run.id)
+        run.request_payload = {
+            "action": "append_update",
+            "provider_payload": dict(payload),
+        }
+        await sync_to_async(run.save)(update_fields=["request_payload", "updated_at"])
+
+        idempotency_key = f"claude_cmd:{trigger.id}:{mode}:{task.id}:{hashlib.sha1(str(body_text or '').encode('utf-8')).hexdigest()[:12]}"
+        await sync_to_async(queue_codex_event_with_pre_approval)(
+            user=trigger.user,
+            run=run,
+            task=task,
+            task_event=None,
+            action="append_update",
+            provider_payload=dict(payload),
+            idempotency_key=idempotency_key,
+        )
+        return CommandResult(
+            ok=True,
+            status="ok",
+            payload={"codex_run_id": str(run.id), "approval_required": True},
+        )
+
+    async def execute(self, ctx: CommandContext) -> CommandResult:
+        trigger = await self._load_trigger(ctx.message_id)
+        if trigger is None:
+            return CommandResult(ok=False, status="failed", error="trigger_not_found")
+
+        profile = await sync_to_async(
+            lambda: CommandProfile.objects.filter(
+                user=trigger.user, slug=self.slug, enabled=True
+            ).first()
+        )()
+        if profile is None:
+            return CommandResult(ok=False, status="skipped", error="profile_missing")
+
+        parsed = parse_claude_command(ctx.message_text)
+        if not parsed.command:
+            return CommandResult(
+                ok=False, status="skipped", error="claude_command_not_matched"
+            )
+
+        service, channel = self._effective_scope(trigger)
+
+        if parsed.command == "status":
+            project = None
+            reply_task = await self._linked_task_from_reply(
+                trigger.user, trigger.reply_to
+            )
+            if reply_task is not None:
+                project = reply_task.project
+            return await self._run_status(trigger, service, channel, project)
+
+        if parsed.command in {"approve", "deny"}:
+            return await self._run_approval_action(
+                trigger,
+                parsed,
+                current_service=str(ctx.service or ""),
+                current_channel=str(ctx.channel_identifier or ""),
+            )
+
+        project_token, cleaned_body = self._extract_project_token(parsed.body_text)
+        reference_code = self._extract_reference(cleaned_body)
+        reply_task = await self._linked_task_from_reply(trigger.user, trigger.reply_to)
+        task = await self._resolve_task(trigger.user, reference_code, reply_task)
+        if task is None:
+            return CommandResult(
+                ok=False, status="failed", error="task_target_required"
+            )
+
+        project, project_error = await self._resolve_project(
+            user=trigger.user,
+            service=service,
+            channel=channel,
+            task=task,
+            reply_task=reply_task,
+            project_token=project_token,
+        )
+        if project is None:
+            return CommandResult(
+                ok=False, status="failed", error=project_error or "project_unresolved"
+            )
+
+        mode = "plan" if parsed.command == "plan" else "default"
+        return await self._create_submission(
+            trigger=trigger,
+            mode=mode,
+            body_text=cleaned_body,
+            task=task,
+            project=project,
+        )
--- a/core/commands/handlers/codex.py
+++ b/core/commands/handlers/codex.py
@@ -0,0 +1,627 @@
+from __future__ import annotations
+
+import hashlib
+import re
+
+from asgiref.sync import sync_to_async
+from django.utils import timezone
+
+from core.commands.base import CommandContext, CommandHandler, CommandResult
+from core.commands.delivery import post_status_in_source
+from core.messaging.text_export import plain_text_blob
+from core.models import (
+    ChatTaskSource,
+    CodexPermissionRequest,
+    CodexRun,
+    CommandProfile,
+    DerivedTask,
+    ExternalSyncEvent,
+    Message,
+    TaskProject,
+    TaskProviderConfig,
+)
+from core.tasks.codex_approval import queue_codex_event_with_pre_approval
+from core.tasks.codex_support import channel_variants, resolve_external_chat_id
+
+_CODEX_DEFAULT_RE = re.compile(
+    r"^\s*(?:\.codex\b|#codex#?)(?P<body>.*)$",
+    re.IGNORECASE | re.DOTALL,
+)
+_CODEX_PLAN_RE = re.compile(
+    r"^\s*(?:\.codex\s+plan\b|#codex\s+plan#?)(?P<body>.*)$",
+    re.IGNORECASE | re.DOTALL,
+)
+_CODEX_STATUS_RE = re.compile(
+    r"^\s*(?:\.codex\s+status\b|#codex\s+status#?)\s*$", re.IGNORECASE
+)
+_CODEX_APPROVE_DENY_RE = re.compile(
+    r"^\s*(?:\.codex|#codex)\s+(?P<action>approve|deny)\s+(?P<approval_key>[A-Za-z0-9._:-]+)#?\s*$",
+    re.IGNORECASE,
+)
+_PROJECT_TOKEN_RE = re.compile(r"\[\s*project\s*:\s*([^\]]+)\]", re.IGNORECASE)
+_REFERENCE_RE = re.compile(r"(?<!\w)#([A-Za-z0-9_-]+)\b")
+
+
+class CodexParsedCommand(dict):
+    @property
+    def command(self) -> str | None:
+        value = self.get("command")
+        return str(value) if value else None
+
+    @property
+    def body_text(self) -> str:
+        return str(self.get("body_text") or "")
+
+    @property
+    def approval_key(self) -> str:
+        return str(self.get("approval_key") or "")
+
+
+def parse_codex_command(text: str) -> CodexParsedCommand:
+    body = str(text or "")
+    m = _CODEX_APPROVE_DENY_RE.match(body)
+    if m:
+        return CodexParsedCommand(
+            command=str(m.group("action") or "").strip().lower(),
+            body_text="",
+            approval_key=str(m.group("approval_key") or "").strip(),
+        )
+    if _CODEX_STATUS_RE.match(body):
+        return CodexParsedCommand(command="status", body_text="", approval_key="")
+    m = _CODEX_PLAN_RE.match(body)
+    if m:
+        return CodexParsedCommand(
+            command="plan",
+            body_text=str(m.group("body") or "").strip(),
+            approval_key="",
+        )
+    m = _CODEX_DEFAULT_RE.match(body)
+    if m:
+        return CodexParsedCommand(
+            command="default",
+            body_text=str(m.group("body") or "").strip(),
+            approval_key="",
+        )
+    return CodexParsedCommand(command=None, body_text="", approval_key="")
+
+
+def codex_trigger_matches(
+    message_text: str, trigger_token: str, exact_match_only: bool
+) -> bool:
+    body = str(message_text or "").strip()
+    parsed = parse_codex_command(body)
+    if parsed.command:
+        return True
+    trigger = str(trigger_token or "").strip()
+    if not trigger:
+        return False
+    if exact_match_only:
+        return body.lower() == trigger.lower()
+    return trigger.lower() in body.lower()
+
+
+class CodexCommandHandler(CommandHandler):
+    slug = "codex"
+
+    async def _load_trigger(self, message_id: str) -> Message | None:
+        return await sync_to_async(
+            lambda: Message.objects.select_related(
+                "user", "session", "session__identifier", "reply_to"
+            )
+            .filter(id=message_id)
+            .first()
+        )()
+
+    def _effective_scope(self, trigger: Message) -> tuple[str, str]:
+        service = str(getattr(trigger, "source_service", "") or "").strip().lower()
+        channel = str(getattr(trigger, "source_chat_id", "") or "").strip()
+        identifier = getattr(getattr(trigger, "session", None), "identifier", None)
+        fallback_service = str(getattr(identifier, "service", "") or "").strip().lower()
+        fallback_identifier = str(getattr(identifier, "identifier", "") or "").strip()
+        if (
+            service == "web"
+            and fallback_service
+            and fallback_identifier
+            and fallback_service != "web"
+        ):
+            return fallback_service, fallback_identifier
+        return service or "web", channel
+
+    async def _mapped_sources(
+        self, user, service: str, channel: str
+    ) -> list[ChatTaskSource]:
+        variants = channel_variants(service, channel)
+        if not variants:
+            return []
+        return await sync_to_async(list)(
+            ChatTaskSource.objects.filter(
+                user=user,
+                enabled=True,
+                service=service,
+                channel_identifier__in=variants,
+            ).select_related("project", "epic")
+        )
+
+    async def _linked_task_from_reply(
+        self, user, reply_to: Message | None
+    ) -> DerivedTask | None:
+        if reply_to is None:
+            return None
+        by_origin = await sync_to_async(
+            lambda: DerivedTask.objects.filter(user=user, origin_message=reply_to)
+            .select_related("project", "epic")
+            .order_by("-created_at")
+            .first()
+        )()
+        if by_origin is not None:
+            return by_origin
+        return await sync_to_async(
+            lambda: DerivedTask.objects.filter(
+                user=user, events__source_message=reply_to
+            )
+            .select_related("project", "epic")
+            .order_by("-created_at")
+            .first()
+        )()
+
+    def _extract_project_token(self, body_text: str) -> tuple[str, str]:
+        text = str(body_text or "")
+        m = _PROJECT_TOKEN_RE.search(text)
+        if not m:
+            return "", text
+        token = str(m.group(1) or "").strip()
+        cleaned = _PROJECT_TOKEN_RE.sub("", text).strip()
+        return token, cleaned
+
+    def _extract_reference(self, body_text: str) -> str:
+        m = _REFERENCE_RE.search(str(body_text or ""))
+        if not m:
+            return ""
+        return str(m.group(1) or "").strip()
+
+    async def _resolve_task(
+        self, user, reference_code: str, reply_task: DerivedTask | None
+    ) -> DerivedTask | None:
+        if reference_code:
+            return await sync_to_async(
+                lambda: DerivedTask.objects.filter(
+                    user=user, reference_code=reference_code
+                )
+                .select_related("project", "epic")
+                .order_by("-created_at")
+                .first()
+            )()
+        return reply_task
+
+    async def _resolve_project(
+        self,
+        *,
+        user,
+        service: str,
+        channel: str,
+        task: DerivedTask | None,
+        reply_task: DerivedTask | None,
+        project_token: str,
+    ) -> tuple[TaskProject | None, str]:
+        if task is not None:
+            return task.project, ""
+        if reply_task is not None:
+            return reply_task.project, ""
+        if project_token:
+            project = await sync_to_async(
+                lambda: TaskProject.objects.filter(
+                    user=user, name__iexact=project_token
+                ).first()
+            )()
+            if project is not None:
+                return project, ""
+            return None, f"project_not_found:{project_token}"
+
+        mapped = await self._mapped_sources(user, service, channel)
+        project_ids = sorted({str(row.project_id) for row in mapped if row.project_id})
+        if len(project_ids) == 1:
+            project = next(
+                (
+                    row.project
+                    for row in mapped
+                    if str(row.project_id) == project_ids[0]
+                ),
+                None,
+            )
+            return project, ""
+        if len(project_ids) > 1:
+            return None, "project_required:[project:Name]"
+        return None, "project_unresolved"
+
+    async def _post_source_status(
+        self, trigger: Message, text: str, suffix: str
+    ) -> None:
+        await post_status_in_source(
+            trigger_message=trigger,
+            text=text,
+            origin_tag=f"codex-status:{suffix}",
+        )
+
+    async def _run_status(
+        self, trigger: Message, service: str, channel: str, project: TaskProject | None
+    ) -> CommandResult:
+        def _load_runs():
+            qs = CodexRun.objects.filter(user=trigger.user)
+            if service:
+                qs = qs.filter(source_service=service)
+            if channel:
+                qs = qs.filter(source_channel=channel)
+            if project is not None:
+                qs = qs.filter(project=project)
+            return list(qs.order_by("-created_at")[:10])
+
+        runs = await sync_to_async(_load_runs)()
+        if not runs:
+            await self._post_source_status(
+                trigger, "[codex] no recent runs for this scope.", "empty"
+            )
+            return CommandResult(ok=True, status="ok", payload={"count": 0})
+        lines = ["[codex] recent runs:"]
+        for row in runs:
+            ref = str(getattr(getattr(row, "task", None), "reference_code", "") or "-")
+            summary = str((row.result_payload or {}).get("summary") or "").strip()
+            summary_part = f" · {summary}" if summary else ""
+            lines.append(f"- {row.status} run={row.id} task=#{ref}{summary_part}")
+        await self._post_source_status(trigger, "\n".join(lines), "runs")
+        return CommandResult(ok=True, status="ok", payload={"count": len(runs)})
+
+    async def _run_approval_action(
+        self,
+        trigger: Message,
+        parsed: CodexParsedCommand,
+        current_service: str,
+        current_channel: str,
+    ) -> CommandResult:
+        cfg = await sync_to_async(
+            lambda: TaskProviderConfig.objects.filter(
+                user=trigger.user, provider="codex_cli"
+            ).first()
+        )()
+        settings_payload = dict(getattr(cfg, "settings", {}) or {})
+        approver_service = (
+            str(settings_payload.get("approver_service") or "").strip().lower()
+        )
+        approver_identifier = str(
+            settings_payload.get("approver_identifier") or ""
+        ).strip()
+        if not approver_service or not approver_identifier:
+            return CommandResult(
+                ok=False, status="failed", error="approver_channel_not_configured"
+            )
+
+        if str(current_service or "").strip().lower() != approver_service or str(
+            current_channel or ""
+        ).strip() not in set(channel_variants(approver_service, approver_identifier)):
+            return CommandResult(
+                ok=False,
+                status="failed",
+                error="approval_command_not_allowed_in_this_channel",
+            )
+
+        approval_key = parsed.approval_key
+        request = await sync_to_async(
+            lambda: CodexPermissionRequest.objects.select_related(
+                "codex_run", "external_sync_event"
+            )
+            .filter(user=trigger.user, approval_key=approval_key)
+            .first()
+        )()
+        if request is None:
+            return CommandResult(
+                ok=False, status="failed", error="approval_key_not_found"
+            )
+
+        now = timezone.now()
+        if parsed.command == "approve":
+            request.status = "approved"
+            request.resolved_at = now
+            request.resolved_by_identifier = current_channel
+            request.resolution_note = "approved via command"
+            await sync_to_async(request.save)(
+                update_fields=[
+                    "status",
+                    "resolved_at",
+                    "resolved_by_identifier",
+                    "resolution_note",
+                ]
+            )
+            if request.external_sync_event_id:
+                await sync_to_async(
+                    ExternalSyncEvent.objects.filter(
+                        id=request.external_sync_event_id
+                    ).update
+                )(
+                    status="ok",
+                    error="",
+                )
+            run = request.codex_run
+            run.status = "approved_waiting_resume"
+            run.error = ""
+            await sync_to_async(run.save)(
+                update_fields=["status", "error", "updated_at"]
+            )
+            source_service = str(run.source_service or "")
+            source_channel = str(run.source_channel or "")
+            resume_payload = dict(request.resume_payload or {})
+            resume_action = str(resume_payload.get("action") or "").strip().lower()
+            resume_provider_payload = dict(resume_payload.get("provider_payload") or {})
+            if resume_action and resume_provider_payload:
+                provider_payload = dict(resume_provider_payload)
+                provider_payload["codex_run_id"] = str(run.id)
+                provider_payload["source_service"] = source_service
+                provider_payload["source_channel"] = source_channel
+                event_action = resume_action
+                resume_idempotency_key = str(
+                    resume_payload.get("idempotency_key") or ""
+                ).strip()
+                resume_event_key = (
+                    resume_idempotency_key
+                    if resume_idempotency_key
+                    else f"codex_approval:{approval_key}:approved"
+                )
+            else:
+                provider_payload = dict(
+                    run.request_payload.get("provider_payload") or {}
+                )
+                provider_payload.update(
+                    {
+                        "mode": "approval_response",
+                        "approval_key": approval_key,
+                        "resume_payload": dict(request.resume_payload or {}),
+                        "codex_run_id": str(run.id),
+                        "source_service": source_service,
+                        "source_channel": source_channel,
+                    }
+                )
+                event_action = "append_update"
+                resume_event_key = f"codex_approval:{approval_key}:approved"
+            await sync_to_async(ExternalSyncEvent.objects.update_or_create)(
+                idempotency_key=resume_event_key,
+                defaults={
+                    "user": trigger.user,
+                    "task_id": run.task_id,
+                    "task_event_id": run.derived_task_event_id,
+                    "provider": "codex_cli",
+                    "status": "pending",
+                    "payload": {
+                        "action": event_action,
+                        "provider_payload": provider_payload,
+                    },
+                    "error": "",
+                },
+            )
+            return CommandResult(
+                ok=True,
+                status="ok",
+                payload={"approval_key": approval_key, "resolution": "approved"},
+            )
+
+        request.status = "denied"
+        request.resolved_at = now
+        request.resolved_by_identifier = current_channel
+        request.resolution_note = "denied via command"
+        await sync_to_async(request.save)(
+            update_fields=[
+                "status",
+                "resolved_at",
+                "resolved_by_identifier",
+                "resolution_note",
+            ]
+        )
+        if request.external_sync_event_id:
+            await sync_to_async(
+                ExternalSyncEvent.objects.filter(
+                    id=request.external_sync_event_id
+                ).update
+            )(
+                status="failed",
+                error="approval_denied",
+            )
+        run = request.codex_run
+        run.status = "denied"
+        run.error = "approval_denied"
+        await sync_to_async(run.save)(update_fields=["status", "error", "updated_at"])
+        await sync_to_async(ExternalSyncEvent.objects.update_or_create)(
+            idempotency_key=f"codex_approval:{approval_key}:denied",
+            defaults={
+                "user": trigger.user,
+                "task_id": run.task_id,
+                "task_event_id": run.derived_task_event_id,
+                "provider": "codex_cli",
+                "status": "failed",
+                "payload": {
+                    "action": "append_update",
+                    "provider_payload": {
+                        "mode": "approval_response",
+                        "approval_key": approval_key,
+                        "codex_run_id": str(run.id),
+                    },
+                },
+                "error": "approval_denied",
+            },
+        )
+        return CommandResult(
+            ok=True,
+            status="ok",
+            payload={"approval_key": approval_key, "resolution": "denied"},
+        )
+
+    async def _create_submission(
+        self,
+        *,
+        trigger: Message,
+        mode: str,
+        body_text: str,
+        task: DerivedTask,
+        project: TaskProject,
+    ) -> CommandResult:
+        cfg = await sync_to_async(
+            lambda: TaskProviderConfig.objects.filter(
+                user=trigger.user, provider="codex_cli", enabled=True
+            ).first()
+        )()
+        if cfg is None:
+            return CommandResult(
+                ok=False, status="failed", error="provider_disabled_or_missing"
+            )
+
+        service, channel = self._effective_scope(trigger)
+        external_chat_id = await sync_to_async(resolve_external_chat_id)(
+            user=trigger.user,
+            provider="codex_cli",
+            service=service,
+            channel=channel,
+        )
+        payload = {
+            "task_id": str(task.id),
+            "reference_code": str(task.reference_code or ""),
+            "title": str(task.title or ""),
+            "external_key": str(task.external_key or ""),
+            "project_name": str(getattr(project, "name", "") or ""),
+            "epic_name": str(getattr(getattr(task, "epic", None), "name", "") or ""),
+            "source_service": service,
+            "source_channel": channel,
+            "external_chat_id": external_chat_id,
+            "origin_message_id": str(getattr(task, "origin_message_id", "") or ""),
+            "trigger_message_id": str(trigger.id),
+            "mode": mode,
+            "command_text": str(body_text or ""),
+        }
+        if mode == "plan":
+            anchor = trigger.reply_to
+            if anchor is None:
+                return CommandResult(
+                    ok=False, status="failed", error="reply_required_for_codex_plan"
+                )
+            rows = await sync_to_async(list)(
+                Message.objects.filter(
+                    user=trigger.user,
+                    session=trigger.session,
+                    ts__gte=int(anchor.ts or 0),
+                    ts__lte=int(trigger.ts or 0),
+                )
+                .order_by("ts")
+                .select_related(
+                    "session", "session__identifier", "session__identifier__person"
+                )
+            )
+            payload["reply_context"] = {
+                "anchor_message_id": str(anchor.id),
+                "trigger_message_id": str(trigger.id),
+                "message_ids": [str(row.id) for row in rows],
+                "content": plain_text_blob(rows),
+            }
+
+        run = await sync_to_async(CodexRun.objects.create)(
+            user=trigger.user,
+            task=task,
+            source_message=trigger,
+            project=project,
+            epic=getattr(task, "epic", None),
+            source_service=service,
+            source_channel=channel,
+            external_chat_id=external_chat_id,
+            status="waiting_approval",
+            request_payload={
+                "action": "append_update",
+                "provider_payload": dict(payload),
+            },
+            result_payload={},
+            error="",
+        )
+        payload["codex_run_id"] = str(run.id)
+        run.request_payload = {
+            "action": "append_update",
+            "provider_payload": dict(payload),
+        }
+        await sync_to_async(run.save)(update_fields=["request_payload", "updated_at"])
+
+        idempotency_key = f"codex_cmd:{trigger.id}:{mode}:{task.id}:{hashlib.sha1(str(body_text or '').encode('utf-8')).hexdigest()[:12]}"
+        await sync_to_async(queue_codex_event_with_pre_approval)(
+            user=trigger.user,
+            run=run,
+            task=task,
+            task_event=None,
+            action="append_update",
+            provider_payload=dict(payload),
+            idempotency_key=idempotency_key,
+        )
+        return CommandResult(
+            ok=True,
+            status="ok",
+            payload={"codex_run_id": str(run.id), "approval_required": True},
+        )
+
+    async def execute(self, ctx: CommandContext) -> CommandResult:
+        trigger = await self._load_trigger(ctx.message_id)
+        if trigger is None:
+            return CommandResult(ok=False, status="failed", error="trigger_not_found")
+
+        profile = await sync_to_async(
+            lambda: CommandProfile.objects.filter(
+                user=trigger.user, slug=self.slug, enabled=True
+            ).first()
+        )()
+        if profile is None:
+            return CommandResult(ok=False, status="skipped", error="profile_missing")
+
+        parsed = parse_codex_command(ctx.message_text)
+        if not parsed.command:
+            return CommandResult(
+                ok=False, status="skipped", error="codex_command_not_matched"
+            )
+
+        service, channel = self._effective_scope(trigger)
+
+        if parsed.command == "status":
+            project = None
+            reply_task = await self._linked_task_from_reply(
+                trigger.user, trigger.reply_to
+            )
+            if reply_task is not None:
+                project = reply_task.project
+            return await self._run_status(trigger, service, channel, project)
+
+        if parsed.command in {"approve", "deny"}:
+            return await self._run_approval_action(
+                trigger,
+                parsed,
+                current_service=str(ctx.service or ""),
+                current_channel=str(ctx.channel_identifier or ""),
+            )
+
+        project_token, cleaned_body = self._extract_project_token(parsed.body_text)
+        reference_code = self._extract_reference(cleaned_body)
+        reply_task = await self._linked_task_from_reply(trigger.user, trigger.reply_to)
+        task = await self._resolve_task(trigger.user, reference_code, reply_task)
+        if task is None:
+            return CommandResult(
+                ok=False, status="failed", error="task_target_required"
+            )
+
+        project, project_error = await self._resolve_project(
+            user=trigger.user,
+            service=service,
+            channel=channel,
+            task=task,
+            reply_task=reply_task,
+            project_token=project_token,
+        )
+        if project is None:
+            return CommandResult(
+                ok=False, status="failed", error=project_error or "project_unresolved"
+            )
+
+        mode = "plan" if parsed.command == "plan" else "default"
+        return await self._create_submission(
+            trigger=trigger,
+            mode=mode,
+            body_text=cleaned_body,
+            task=task,
+            project=project,
+        )
--- a/core/commands/policies.py
+++ b/core/commands/policies.py
@@ -0,0 +1,111 @@
+from __future__ import annotations
+
+from typing import Iterable
+
+from core.models import CommandAction, CommandProfile, CommandVariantPolicy
+
+BP_VARIANT_KEYS = ("bp", "bp_set", "bp_set_range")
+BP_VARIANT_META = {
+    "bp": {
+        "name": "bp",
+        "trigger_token": ".bp",
+        "template_supported": True,
+        "position": 0,
+    },
+    "bp_set": {
+        "name": "bp set",
+        "trigger_token": ".bp set",
+        "template_supported": False,
+        "position": 1,
+    },
+    "bp_set_range": {
+        "name": "bp set range",
+        "trigger_token": ".bp set range",
+        "template_supported": False,
+        "position": 2,
+    },
+}
+
+
+def _legacy_defaults(profile: CommandProfile, post_result_enabled: bool) -> dict:
+    return {
+        "enabled": True,
+        "generation_mode": "ai",
+        "send_plan_to_egress": bool(post_result_enabled),
+        "send_status_to_source": str(profile.visibility_mode or "")
+        == "status_in_source",
+        "send_status_to_egress": False,
+        "store_document": True,
+    }
+
+
+def _bp_defaults(
+    profile: CommandProfile,
+    variant_key: str,
+    post_result_enabled: bool,
+) -> dict:
+    defaults = _legacy_defaults(profile, post_result_enabled)
+    if variant_key in {"bp_set", "bp_set_range"}:
+        defaults["generation_mode"] = "verbatim"
+    else:
+        defaults["generation_mode"] = "ai"
+    return defaults
+
+
+def ensure_variant_policies_for_profile(
+    profile: CommandProfile,
+    *,
+    action_rows: Iterable[CommandAction] | None = None,
+) -> dict[str, CommandVariantPolicy]:
+    actions = (
+        list(action_rows) if action_rows is not None else list(profile.actions.all())
+    )
+    post_result_enabled = any(
+        row.action_type == "post_result" and bool(row.enabled) for row in actions
+    )
+    result: dict[str, CommandVariantPolicy] = {}
+
+    if str(profile.slug or "").strip() == "bp":
+        for key in BP_VARIANT_KEYS:
+            meta = BP_VARIANT_META.get(key, {})
+            defaults = _bp_defaults(profile, key, post_result_enabled)
+            policy, _ = CommandVariantPolicy.objects.get_or_create(
+                profile=profile,
+                variant_key=key,
+                defaults={
+                    **defaults,
+                    "position": int(meta.get("position") or 0),
+                },
+            )
+            result[key] = policy
+    else:
+        defaults = _legacy_defaults(profile, post_result_enabled)
+        policy, _ = CommandVariantPolicy.objects.get_or_create(
+            profile=profile,
+            variant_key="default",
+            defaults={
+                **defaults,
+                "generation_mode": "verbatim",
+                "position": 0,
+            },
+        )
+        result["default"] = policy
+
+    return result
+
+
+def load_variant_policy(
+    profile: CommandProfile, variant_key: str
+) -> CommandVariantPolicy | None:
+    key = str(variant_key or "").strip()
+    if not key:
+        return None
+    policy = (
+        profile.variant_policies.filter(variant_key=key)
+        .order_by("position", "id")
+        .first()
+    )
+    if policy is not None:
+        return policy
+    ensured = ensure_variant_policies_for_profile(profile)
+    return ensured.get(key)
--- a/core/commands/registry.py
+++ b/core/commands/registry.py
@@ -0,0 +1,16 @@
+from __future__ import annotations
+
+from core.commands.base import CommandHandler
+
+_HANDLERS: dict[str, CommandHandler] = {}
+
+
+def register(handler: CommandHandler):
+    slug = str(getattr(handler, "slug", "") or "").strip().lower()
+    if not slug:
+        raise ValueError("handler slug is required")
+    _HANDLERS[slug] = handler
+
+
+def get(slug: str) -> CommandHandler | None:
+    return _HANDLERS.get(str(slug or "").strip().lower())
--- a/core/context_processors.py
+++ b/core/context_processors.py
@@ -0,0 +1,165 @@
+from django.urls import reverse
+
+
+def _tab(label: str, href: str, active: bool) -> dict:
+    return {
+        "label": label,
+        "href": href,
+        "active": bool(active),
+    }
+
+
+def settings_hierarchy_nav(request):
+    match = getattr(request, "resolver_match", None)
+    if match is None:
+        return {}
+
+    url_name = str(getattr(match, "url_name", "") or "")
+    namespace = str(getattr(match, "namespace", "") or "")
+    path = str(getattr(request, "path", "") or "")
+
+    notifications_href = reverse("notifications_settings")
+    system_href = reverse("system_settings")
+    accessibility_href = reverse("accessibility_settings")
+    encryption_href = reverse("encryption_settings")
+    permissions_href = reverse("permission_settings")
+    security_2fa_href = reverse("security_2fa")
+    ai_models_href = reverse("ai_models")
+    ai_traces_href = reverse("ai_execution_log")
+    commands_href = reverse("command_routing")
+    business_plans_href = reverse("business_plan_inbox")
+    tasks_href = reverse("tasks_settings")
+    translation_href = reverse("translation_settings")
+    behavioral_href = reverse("behavioral_signals_settings")
+
+    categories = {
+        "general": {
+            "routes": {
+                "notifications_settings",
+                "notifications_update",
+                "system_settings",
+                "accessibility_settings",
+            },
+            "title": "General",
+            "tabs": [
+                (
+                    "Notifications",
+                    notifications_href,
+                    lambda: path == notifications_href,
+                ),
+                ("System", system_href, lambda: path == system_href),
+                (
+                    "Accessibility",
+                    accessibility_href,
+                    lambda: path == accessibility_href,
+                ),
+            ],
+        },
+        "security": {
+            "routes": {
+                "security_settings",
+                "encryption_settings",
+                "permission_settings",
+                "security_2fa",
+            },
+            "title": "Security",
+            "tabs": [
+                ("Encryption", encryption_href, lambda: path == encryption_href),
+                ("Permissions", permissions_href, lambda: path == permissions_href),
+                (
+                    "2FA",
+                    security_2fa_href,
+                    lambda: path == security_2fa_href or namespace == "two_factor",
+                ),
+            ],
+        },
+        "ai": {
+            "routes": {
+                "ai_settings",
+                "ai_models",
+                "ais",
+                "ai_create",
+                "ai_update",
+                "ai_delete",
+                "ai_execution_log",
+            },
+            "title": "AI",
+            "tabs": [
+                ("Models", ai_models_href, lambda: path == ai_models_href),
+                ("Traces", ai_traces_href, lambda: path == ai_traces_href),
+            ],
+        },
+        "modules": {
+            "routes": {
+                "modules_settings",
+                "command_routing",
+                "business_plan_inbox",
+                "business_plan_editor",
+                "tasks_settings",
+                "translation_settings",
+                "translation_preview",
+                "availability_settings",
+                "behavioral_signals_settings",
+                "codex_settings",
+                "codex_approval",
+            },
+            "title": "Modules",
+            "tabs": [
+                ("Commands", commands_href, lambda: path == commands_href),
+                (
+                    "Business Plans",
+                    business_plans_href,
+                    lambda: url_name in {"business_plan_inbox", "business_plan_editor"},
+                ),
+                ("Task Automation", tasks_href, lambda: path == tasks_href),
+                (
+                    "Translation",
+                    translation_href,
+                    lambda: url_name in {"translation_settings", "translation_preview"},
+                ),
+                (
+                    "Behavioral Signals",
+                    behavioral_href,
+                    lambda: url_name
+                    in {"availability_settings", "behavioral_signals_settings"},
+                ),
+            ],
+        },
+    }
+
+    two_factor_security_routes = {
+        "profile",
+        "setup",
+        "backup_tokens",
+        "disable",
+        "phone_create",
+        "phone_delete",
+    }
+
+    if url_name in categories["general"]["routes"]:
+        category = categories["general"]
+    elif url_name in categories["security"]["routes"] or (
+        namespace == "two_factor" and url_name in two_factor_security_routes
+    ):
+        category = categories["security"]
+    elif url_name in categories["ai"]["routes"]:
+        category = categories["ai"]
+    elif url_name in categories["modules"]["routes"]:
+        category = categories["modules"]
+    else:
+        category = None
+
+    if category is None:
+        settings_nav = None
+    else:
+        settings_nav = {
+            "title": str(category.get("title") or "Settings"),
+            "tabs": [
+                _tab(label, href, bool(is_active()))
+                for label, href, is_active in category.get("tabs", [])
+            ],
+        }
+
+    if not settings_nav:
+        return {}
+    return {"settings_nav": settings_nav}
--- a/core/db/sql.py
+++ b/core/db/sql.py
@@ -24,7 +24,6 @@ async def init_mysql_pool():

 async def close_mysql_pool():
    """Close the MySQL connection pool properly."""
-    global mysql_pool
    if mysql_pool:
        mysql_pool.close()
        await mysql_pool.wait_closed()
--- a/core/events/init.py
+++ b/core/events/init.py
@@ -0,0 +1,16 @@
+from core.events.ledger import (
+    append_event,
+    append_event_sync,
+    event_ledger_enabled,
+    event_ledger_status,
+)
+from core.events.projection import project_session_from_events, shadow_compare_session
+
+__all__ = [
+    "append_event",
+    "append_event_sync",
+    "event_ledger_enabled",
+    "event_ledger_status",
+    "project_session_from_events",
+    "shadow_compare_session",
+]
--- a/core/events/behavior.py
+++ b/core/events/behavior.py
@@ -0,0 +1,213 @@
+from __future__ import annotations
+
+import json
+import statistics
+from dataclasses import dataclass
+from typing import Any
+
+
+def safe_int(value: Any, default: int = 0) -> int:
+    try:
+        return int(value)
+    except Exception:
+        return int(default)
+
+
+def parse_payload(value: Any) -> dict:
+    if isinstance(value, dict):
+        return dict(value)
+    if isinstance(value, str):
+        text = value.strip()
+        if not text:
+            return {}
+        try:
+            loaded = json.loads(text)
+        except Exception:
+            return {}
+        if isinstance(loaded, dict):
+            return dict(loaded)
+    return {}
+
+
+def median_ms(values: list[int]) -> int:
+    clean = [int(v) for v in values if safe_int(v, 0) > 0]
+    if not clean:
+        return 0
+    return int(statistics.median(clean))
+
+
+def z_score(value: int, baseline_samples: list[int]) -> float:
+    clean = [int(v) for v in baseline_samples if safe_int(v, 0) > 0]
+    if len(clean) < 2:
+        return 0.0
+    baseline = statistics.median(clean)
+    stdev = statistics.pstdev(clean)
+    if stdev <= 0:
+        return 0.0
+    return float((float(value) - float(baseline)) / float(stdev))
+
+
+@dataclass
+class CompositionState:
+    started_ts: int
+    last_started_ts: int
+    stopped_ts: int = 0
+    revision: int = 1
+
+
+class ComposingTracker:
+    def __init__(self, window_ms: int = 300000):
+        self.window_ms = max(1000, int(window_ms or 300000))
+        self._state: dict[str, CompositionState] = {}
+
+    def observe_started(self, session_id: str, ts: int) -> CompositionState:
+        key = str(session_id or "").strip()
+        if not key:
+            raise ValueError("session_id is required")
+        safe_ts_value = max(0, safe_int(ts, 0))
+        state = self._state.get(key)
+        if state is None:
+            state = CompositionState(
+                started_ts=safe_ts_value,
+                last_started_ts=safe_ts_value,
+                revision=1,
+            )
+            self._state[key] = state
+            return state
+        if state.stopped_ts > 0:
+            state.revision += 1
+        state.last_started_ts = safe_ts_value
+        state.stopped_ts = 0
+        return state
+
+    def observe_stopped(self, session_id: str, ts: int) -> dict | None:
+        key = str(session_id or "").strip()
+        state = self._state.get(key)
+        if state is None:
+            return None
+        safe_ts_value = max(0, safe_int(ts, 0))
+        duration_ms = max(0, safe_ts_value - int(state.started_ts or 0))
+        if duration_ms >= self.window_ms:
+            self._state.pop(key, None)
+            return {
+                "started_ts": int(state.started_ts or 0),
+                "stopped_ts": safe_ts_value,
+                "duration_ms": duration_ms,
+                "revision": int(state.revision or 1),
+                "abandoned": True,
+            }
+        state.stopped_ts = safe_ts_value
+        return None
+
+    def observe_message(self, session_id: str) -> CompositionState | None:
+        key = str(session_id or "").strip()
+        if not key:
+            return None
+        return self._state.pop(key, None)
+
+
+def extract_metric_samples(rows: list[dict]) -> dict[str, list[int]]:
+    delivered_by_message: dict[str, int] = {}
+    read_by_message: dict[str, int] = {}
+    delay_c_samples: list[int] = []
+    delay_f_samples: list[int] = []
+    revision_samples: list[int] = []
+    abandoned_started = 0
+    abandoned_total = 0
+    composition_by_session: dict[str, dict[str, int]] = {}
+    presence_by_session: dict[str, int] = {}
+
+    for row in sorted(
+        list(rows or []),
+        key=lambda item: (
+            safe_int(item.get("ts"), 0),
+            str(item.get("kind") or ""),
+            str(item.get("session_id") or ""),
+        ),
+    ):
+        kind = str(row.get("kind") or "").strip().lower()
+        session_id = str(row.get("session_id") or "").strip()
+        ts = safe_int(row.get("ts"), 0)
+        payload = parse_payload(row.get("payload"))
+        message_id = str(
+            payload.get("message_id")
+            or payload.get("origin_message_id")
+            or row.get("origin_message_id")
+            or ""
+        ).strip()
+
+        if kind == "message_delivered" and message_id:
+            delivered_by_message[message_id] = ts
+            continue
+        if kind == "message_read" and message_id:
+            read_by_message[message_id] = ts
+            continue
+        if kind == "presence_available" and session_id:
+            presence_by_session[session_id] = ts
+            continue
+        if kind == "composing_started" and session_id:
+            abandoned_started += 1
+            state = composition_by_session.get(session_id)
+            if state is None:
+                state = {"started_ts": ts, "revision": 1}
+                composition_by_session[session_id] = state
+            else:
+                state["revision"] = int(state.get("revision", 1)) + 1
+            if presence_by_session.get(session_id):
+                delta = ts - int(presence_by_session.get(session_id) or 0)
+                if delta >= 0:
+                    delay_f_samples.append(delta)
+            continue
+        if kind == "composing_abandoned":
+            abandoned_total += 1
+            if session_id:
+                composition_by_session.pop(session_id, None)
+            continue
+        if kind == "message_sent" and session_id:
+            state = composition_by_session.pop(session_id, None)
+            if state is None:
+                continue
+            delta = ts - int(state.get("started_ts") or 0)
+            if delta >= 0:
+                delay_c_samples.append(delta)
+            revision_samples.append(max(1, int(state.get("revision") or 1)))
+
+    delay_b_samples = []
+    for message_id, delivered_ts in delivered_by_message.items():
+        read_ts = safe_int(read_by_message.get(message_id), 0)
+        if read_ts > 0 and read_ts >= delivered_ts:
+            delay_b_samples.append(read_ts - delivered_ts)
+
+    abandoned_rate_samples = []
+    if abandoned_started > 0:
+        abandoned_rate_samples.append(
+            int(round((float(abandoned_total) / float(abandoned_started)) * 1000))
+        )
+
+    return {
+        "delay_b": delay_b_samples,
+        "delay_c": delay_c_samples,
+        "delay_f": delay_f_samples,
+        "revision": revision_samples,
+        "abandoned_rate": abandoned_rate_samples,
+    }
+
+
+def summarize_metrics(window_rows: list[dict], baseline_rows: list[dict]) -> dict[str, dict]:
+    window_samples = extract_metric_samples(window_rows)
+    baseline_samples = extract_metric_samples(baseline_rows)
+    metrics: dict[str, dict] = {}
+    for metric in ("delay_b", "delay_c", "delay_f", "revision", "abandoned_rate"):
+        samples = list(window_samples.get(metric) or [])
+        if not samples:
+            continue
+        baseline = list(baseline_samples.get(metric) or [])
+        value = median_ms(samples)
+        baseline_value = median_ms(baseline)
+        metrics[metric] = {
+            "value_ms": int(value),
+            "baseline_ms": int(baseline_value),
+            "z_score": float(round(z_score(value, baseline), 6)),
+            "sample_n": len(samples),
+        }
+    return metrics
--- a/core/events/ledger.py
+++ b/core/events/ledger.py
@@ -0,0 +1,157 @@
+from __future__ import annotations
+
+import time
+
+from asgiref.sync import sync_to_async
+from django.conf import settings
+
+from core.events.manticore import get_event_ledger_backend
+from core.models import ConversationEvent
+from core.observability.tracing import ensure_trace_id
+from core.util import logs
+
+log = logs.get_logger("event-ledger")
+
+
+def event_ledger_enabled() -> bool:
+    return bool(
+        getattr(settings, "EVENT_LEDGER_DUAL_WRITE", False)
+        or getattr(settings, "EVENT_PRIMARY_WRITE_PATH", False)
+    )
+
+
+def event_ledger_status() -> dict:
+    return {
+        "event_ledger_dual_write": bool(
+            getattr(settings, "EVENT_LEDGER_DUAL_WRITE", False)
+        ),
+        "event_primary_write_path": bool(
+            getattr(settings, "EVENT_PRIMARY_WRITE_PATH", False)
+        ),
+    }
+
+
+def _normalize_direction(value: str) -> str:
+    direction = str(value or "system").strip().lower()
+    if direction not in {"in", "out", "system"}:
+        return "system"
+    return direction
+
+
+def _safe_ts(value: int | None) -> int:
+    if value is None:
+        return int(time.time() * 1000)
+    try:
+        parsed = int(value)
+    except Exception:
+        return int(time.time() * 1000)
+    if parsed <= 0:
+        return int(time.time() * 1000)
+    return parsed
+
+
+def append_event_sync(
+    *,
+    user,
+    session,
+    event_type: str,
+    direction: str,
+    actor_identifier: str = "",
+    origin_transport: str = "",
+    origin_message_id: str = "",
+    origin_chat_id: str = "",
+    payload: dict | None = None,
+    raw_payload: dict | None = None,
+    trace_id: str = "",
+    ts: int | None = None,
+):
+    if not event_ledger_enabled():
+        return None
+
+    normalized_type = str(event_type or "").strip().lower()
+    if not normalized_type:
+        raise ValueError("event_type is required")
+
+    candidates = {str(choice[0]) for choice in ConversationEvent.EVENT_TYPE_CHOICES}
+    if normalized_type not in candidates:
+        raise ValueError(f"unsupported event_type: {normalized_type}")
+
+    normalized_direction = _normalize_direction(direction)
+    normalized_trace = ensure_trace_id(trace_id, payload or {})
+
+    safe_ts = _safe_ts(ts)
+    transport = str(origin_transport or "").strip().lower()
+    message_id = str(origin_message_id or "").strip()
+    actor_identifier = str(actor_identifier or "").strip()
+    origin_chat_id = str(origin_chat_id or "").strip()
+    payload = dict(payload or {})
+    raw_payload = dict(raw_payload or {})
+
+    dual_write = bool(getattr(settings, "EVENT_LEDGER_DUAL_WRITE", False))
+    primary_write = bool(getattr(settings, "EVENT_PRIMARY_WRITE_PATH", False))
+    write_django = dual_write and not primary_write
+
+    row = None
+    if write_django:
+        dedup_row = None
+        if transport and message_id:
+            dedup_row = (
+                ConversationEvent.objects.filter(
+                    user=user,
+                    session=session,
+                    event_type=normalized_type,
+                    origin_transport=transport,
+                    origin_message_id=message_id,
+                )
+                .order_by("-created_at")
+                .first()
+            )
+        if dedup_row is not None:
+            row = dedup_row
+        else:
+            row = ConversationEvent.objects.create(
+                user=user,
+                session=session,
+                ts=safe_ts,
+                event_type=normalized_type,
+                direction=normalized_direction,
+                actor_identifier=actor_identifier,
+                origin_transport=transport,
+                origin_message_id=message_id,
+                origin_chat_id=origin_chat_id,
+                payload=payload,
+                raw_payload=raw_payload,
+                trace_id=normalized_trace,
+            )
+
+    try:
+        get_event_ledger_backend().upsert_event(
+            user_id=int(user.id),
+            person_id=str(session.identifier.person_id),
+            session_id=str(session.id),
+            event_type=normalized_type,
+            direction=normalized_direction,
+            ts=safe_ts,
+            actor_identifier=actor_identifier,
+            origin_transport=transport,
+            origin_message_id=message_id,
+            origin_chat_id=origin_chat_id,
+            payload=payload,
+            raw_payload=raw_payload,
+            trace_id=normalized_trace,
+        )
+    except Exception as exc:
+        if primary_write:
+            raise
+        log.warning(
+            "Event ledger manticore dual-write failed session=%s event_type=%s err=%s",
+            getattr(session, "id", "-"),
+            normalized_type,
+            exc,
+        )
+
+    return row
+
+
+async def append_event(**kwargs):
+    return await sync_to_async(append_event_sync, thread_sensitive=True)(**kwargs)
--- a/core/events/manticore.py
+++ b/core/events/manticore.py
@@ -0,0 +1,588 @@
+from __future__ import annotations
+
+import hashlib
+import json
+import time
+from urllib.parse import urlparse, urlunparse
+from typing import Any
+
+import requests
+from django.conf import settings
+
+from core.models import ConversationEvent
+from core.util import logs
+from core.events.behavior import parse_payload
+
+log = logs.get_logger("event-manticore")
+
+
+class ManticoreEventLedgerBackend:
+    _table_ready_cache: dict[str, float] = {}
+    _table_ready_ttl_seconds = 30.0
+
+    def __init__(self):
+        self.base_url = str(
+            getattr(settings, "MANTICORE_HTTP_URL", "http://localhost:9308")
+        ).rstrip("/")
+        self.table = (
+            str(getattr(settings, "MANTICORE_EVENT_TABLE", "gia_events")).strip()
+            or "gia_events"
+        )
+        self.metrics_table = (
+            str(getattr(settings, "MANTICORE_METRIC_TABLE", "gia_metrics")).strip()
+            or "gia_metrics"
+        )
+        self.timeout_seconds = int(getattr(settings, "MANTICORE_HTTP_TIMEOUT", 5) or 5)
+        self._table_cache_key = f"{self.base_url}|{self.table}"
+        self._metrics_cache_key = f"{self.base_url}|{self.metrics_table}"
+
+    def _candidate_base_urls(self) -> list[str]:
+        parsed = urlparse(self.base_url)
+        hostname = str(parsed.hostname or "").strip().lower()
+        candidates = [self.base_url]
+        if hostname in {"localhost", "127.0.0.1"}:
+            replacement = parsed._replace(netloc=f"host.containers.internal:{parsed.port or 9308}")
+            candidates.append(urlunparse(replacement))
+        output = []
+        seen = set()
+        for value in candidates:
+            key = str(value or "").strip()
+            if not key or key in seen:
+                continue
+            seen.add(key)
+            output.append(key)
+        return output
+
+    def _sql(self, query: str) -> dict[str, Any]:
+        last_exc = None
+        for base_url in self._candidate_base_urls():
+            try:
+                response = requests.post(
+                    f"{base_url}/sql",
+                    data={"mode": "raw", "query": query},
+                    timeout=self.timeout_seconds,
+                )
+                response.raise_for_status()
+                payload = response.json()
+                if base_url != self.base_url:
+                    self.base_url = base_url.rstrip("/")
+                    self._table_cache_key = f"{self.base_url}|{self.table}"
+                    self._metrics_cache_key = f"{self.base_url}|{self.metrics_table}"
+                if isinstance(payload, list):
+                    return payload[0] if payload else {}
+                return dict(payload or {})
+            except Exception as exc:
+                last_exc = exc
+        if last_exc is not None:
+            raise last_exc
+        return {}
+
+    def ensure_table(self) -> None:
+        last_ready = float(
+            self._table_ready_cache.get(self._table_cache_key, 0.0) or 0.0
+        )
+        if (time.time() - last_ready) <= float(self._table_ready_ttl_seconds):
+            return
+        self._sql(
+            (
+                f"CREATE TABLE IF NOT EXISTS {self.table} ("
+                "id BIGINT,"
+                "user_id BIGINT,"
+                "person_id STRING,"
+                "session_id STRING,"
+                "transport STRING,"
+                "kind STRING,"
+                "direction STRING,"
+                "ts BIGINT,"
+                "ts_ref BIGINT,"
+                "actor STRING,"
+                "duration_ms BIGINT,"
+                "abandoned INTEGER,"
+                "revision INTEGER,"
+                "payload JSON"
+                ") engine='columnar' min_infix_len='2'"
+            )
+        )
+        self._table_ready_cache[self._table_cache_key] = time.time()
+
+    def ensure_metrics_table(self) -> None:
+        last_ready = float(
+            self._table_ready_cache.get(self._metrics_cache_key, 0.0) or 0.0
+        )
+        if (time.time() - last_ready) <= float(self._table_ready_ttl_seconds):
+            return
+        self._sql(
+            (
+                f"CREATE TABLE IF NOT EXISTS {self.metrics_table} ("
+                "id BIGINT,"
+                "user_id BIGINT,"
+                "person_id STRING,"
+                "window_days INTEGER,"
+                "metric STRING,"
+                "value_ms BIGINT,"
+                "baseline_ms BIGINT,"
+                "z_score FLOAT,"
+                "sample_n INTEGER,"
+                "computed_at BIGINT"
+                ") engine='columnar'"
+            )
+        )
+        self._table_ready_cache[self._metrics_cache_key] = time.time()
+
+    def _escape(self, value: Any) -> str:
+        text = str(value or "")
+        return text.replace("\\", "\\\\").replace("'", "\\'")
+
+    def _event_id(self, *, logical_key: str) -> int:
+        digest = hashlib.blake2b(
+            str(logical_key or "").encode("utf-8"),
+            digest_size=8,
+        ).digest()
+        value = int.from_bytes(digest, byteorder="big", signed=False)
+        return max(1, int(value))
+
+    def _event_kind(self, event_type: str) -> str:
+        normalized = str(event_type or "").strip().lower()
+        return {
+            "message_created": "message_sent",
+            "delivery_receipt": "message_delivered",
+            "read_receipt": "message_read",
+            "typing_started": "composing_started",
+            "typing_stopped": "composing_stopped",
+            "composing_abandoned": "composing_abandoned",
+            "presence_available": "presence_available",
+            "presence_unavailable": "presence_unavailable",
+        }.get(normalized, normalized)
+
+    def _rows_from_sql_payload(self, payload: dict[str, Any]) -> list[dict]:
+        data = payload.get("data") or payload.get("hits") or []
+        if isinstance(data, dict):
+            data = [data]
+        rows = []
+        for row in list(data or []):
+            if isinstance(row, dict):
+                rows.append(dict(row))
+        return rows
+
+    def _build_values(
+        self,
+        *,
+        user_id: int,
+        person_id: str,
+        session_id: str,
+        event_type: str,
+        direction: str,
+        ts: int,
+        actor_identifier: str,
+        origin_transport: str,
+        origin_message_id: str,
+        origin_chat_id: str,
+        payload: dict | None,
+        raw_payload: dict | None,
+        trace_id: str,
+    ) -> str:
+        data = dict(payload or {})
+        if raw_payload:
+            data["raw_payload"] = dict(raw_payload)
+        if trace_id:
+            data["trace_id"] = str(trace_id)
+        if origin_message_id:
+            data["origin_message_id"] = str(origin_message_id)
+        if origin_chat_id:
+            data["origin_chat_id"] = str(origin_chat_id)
+        data["legacy_event_type"] = str(event_type or "").strip().lower()
+
+        ts_ref = 0
+        try:
+            ts_ref = int(data.get("message_ts") or data.get("source_ts") or 0)
+        except Exception:
+            ts_ref = 0
+        try:
+            duration_ms = int(data.get("duration_ms") or 0)
+        except Exception:
+            duration_ms = 0
+        try:
+            abandoned = 1 if bool(data.get("abandoned")) else 0
+        except Exception:
+            abandoned = 0
+        try:
+            revision = int(data.get("revision") or 0)
+        except Exception:
+            revision = 0
+
+        logical_key = "|".join(
+            [
+                str(user_id),
+                str(session_id),
+                str(event_type or "").strip().lower(),
+                str(direction or "").strip().lower(),
+                str(origin_transport or "").strip().lower(),
+                str(origin_message_id or "").strip(),
+                str(origin_chat_id or "").strip(),
+                str(actor_identifier or "").strip(),
+                str(int(ts or 0)),
+                str(trace_id or "").strip(),
+            ]
+        )
+        doc_id = self._event_id(logical_key=logical_key)
+        payload_json = json.dumps(data, separators=(",", ":"), sort_keys=True)
+        return (
+            f"({doc_id},{int(user_id)},'{self._escape(person_id)}',"
+            f"'{self._escape(session_id)}','{self._escape(origin_transport)}',"
+            f"'{self._escape(self._event_kind(event_type))}','{self._escape(direction)}',"
+            f"{int(ts)},{ts_ref},'{self._escape(actor_identifier)}',{duration_ms},"
+            f"{abandoned},{revision},'{self._escape(payload_json)}')"
+        )
+
+    def upsert_event(
+        self,
+        *,
+        user_id: int,
+        person_id: str,
+        session_id: str,
+        event_type: str,
+        direction: str,
+        ts: int,
+        actor_identifier: str = "",
+        origin_transport: str = "",
+        origin_message_id: str = "",
+        origin_chat_id: str = "",
+        payload: dict | None = None,
+        raw_payload: dict | None = None,
+        trace_id: str = "",
+    ) -> None:
+        self.ensure_table()
+        values = self._build_values(
+            user_id=user_id,
+            person_id=person_id,
+            session_id=session_id,
+            event_type=event_type,
+            direction=direction,
+            ts=ts,
+            actor_identifier=actor_identifier,
+            origin_transport=origin_transport,
+            origin_message_id=origin_message_id,
+            origin_chat_id=origin_chat_id,
+            payload=payload,
+            raw_payload=raw_payload,
+            trace_id=trace_id,
+        )
+        self._sql(
+            f"REPLACE INTO {self.table} "
+            "(id,user_id,person_id,session_id,transport,kind,direction,ts,ts_ref,actor,duration_ms,abandoned,revision,payload) "
+            f"VALUES {values}"
+        )
+
+    def query_rows(self, query: str) -> list[dict]:
+        return self._rows_from_sql_payload(self._sql(query))
+
+    def list_event_targets(self, *, user_id: int | None = None) -> list[dict]:
+        filters = []
+        if user_id is not None:
+            filters.append(f"user_id={int(user_id)}")
+        where_clause = f" WHERE {' AND '.join(filters)}" if filters else ""
+        return self.query_rows(
+            f"SELECT user_id, person_id FROM {self.table}{where_clause} "
+            "GROUP BY user_id, person_id"
+        )
+
+    def fetch_events(
+        self,
+        *,
+        user_id: int,
+        person_id: str,
+        since_ts: int,
+    ) -> list[dict]:
+        return self.query_rows(
+            f"SELECT user_id, person_id, session_id, transport, kind, direction, ts, ts_ref, actor, duration_ms, abandoned, revision, payload "
+            f"FROM {self.table} "
+            f"WHERE user_id={int(user_id)} "
+            f"AND person_id='{self._escape(person_id)}' "
+            f"AND ts>={int(since_ts)} "
+            "ORDER BY ts ASC"
+        )
+
+    def _metric_doc_id(
+        self,
+        *,
+        user_id: int,
+        person_id: str,
+        window_days: int,
+        metric: str,
+    ) -> int:
+        digest = hashlib.blake2b(
+            f"{int(user_id)}|{person_id}|{int(window_days)}|{metric}".encode("utf-8"),
+            digest_size=8,
+        ).digest()
+        return max(1, int.from_bytes(digest, byteorder="big", signed=False))
+
+    def upsert_metric(
+        self,
+        *,
+        user_id: int,
+        person_id: str,
+        window_days: int,
+        metric: str,
+        value_ms: int,
+        baseline_ms: int,
+        z_score: float,
+        sample_n: int,
+        computed_at: int,
+    ) -> None:
+        self.ensure_metrics_table()
+        doc_id = self._metric_doc_id(
+            user_id=user_id,
+            person_id=person_id,
+            window_days=window_days,
+            metric=metric,
+        )
+        self._sql(
+            f"REPLACE INTO {self.metrics_table} "
+            "(id,user_id,person_id,window_days,metric,value_ms,baseline_ms,z_score,sample_n,computed_at) "
+            f"VALUES ({doc_id},{int(user_id)},'{self._escape(person_id)}',{int(window_days)},"
+            f"'{self._escape(metric)}',{int(value_ms)},{int(baseline_ms)},"
+            f"{float(z_score)},{int(sample_n)},{int(computed_at)})"
+        )
+
+
+def get_event_ledger_backend() -> ManticoreEventLedgerBackend:
+    return ManticoreEventLedgerBackend()
+
+
+def upsert_conversation_event(event: ConversationEvent) -> None:
+    session = event.session
+    identifier = session.identifier
+    get_event_ledger_backend().upsert_event(
+        user_id=int(event.user_id),
+        person_id=str(identifier.person_id),
+        session_id=str(session.id),
+        event_type=str(event.event_type or ""),
+        direction=str(event.direction or "system"),
+        ts=int(event.ts or 0),
+        actor_identifier=str(event.actor_identifier or ""),
+        origin_transport=str(event.origin_transport or ""),
+        origin_message_id=str(event.origin_message_id or ""),
+        origin_chat_id=str(event.origin_chat_id or ""),
+        payload=dict(event.payload or {}),
+        raw_payload=dict(event.raw_payload or {}),
+        trace_id=str(event.trace_id or ""),
+    )
+
+
+def get_behavioral_availability_stats(*, user_id: int) -> list[dict]:
+    backend = get_event_ledger_backend()
+    return backend.query_rows(
+        f"SELECT person_id, transport, "
+        "COUNT(*) AS total_events, "
+        "SUM(IF(kind IN ('presence_available','presence_unavailable'),1,0)) AS presence_events, "
+        "SUM(IF(kind='message_read',1,0)) AS read_events, "
+        "SUM(IF(kind IN ('composing_started','composing_stopped'),1,0)) AS typing_events, "
+        "SUM(IF(kind='message_sent',1,0)) AS message_events, "
+        "SUM(IF(kind='composing_abandoned',1,0)) AS abandoned_events, "
+        "MAX(ts) AS last_event_ts "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)} "
+        "GROUP BY person_id, transport "
+        "ORDER BY total_events DESC, person_id ASC, transport ASC"
+    )
+
+
+def get_behavioral_latest_states(
+    *,
+    user_id: int,
+    person_ids: list[str],
+    transport: str = "",
+) -> list[dict]:
+    backend = get_event_ledger_backend()
+    cleaned_ids = [
+        str(value or "").strip()
+        for value in list(person_ids or [])
+        if str(value or "").strip()
+    ]
+    if not cleaned_ids:
+        return []
+    id_clause = ",".join(f"'{backend._escape(value)}'" for value in cleaned_ids)
+    transport_clause = ""
+    if str(transport or "").strip():
+        transport_clause = (
+            f" AND transport='{backend._escape(str(transport or '').strip().lower())}'"
+        )
+    return backend.query_rows(
+        f"SELECT person_id, transport, kind, ts "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)} "
+        f"AND person_id IN ({id_clause})"
+        f"{transport_clause} "
+        "ORDER BY person_id ASC, ts DESC"
+    )
+
+
+def get_behavioral_events_for_range(
+    *,
+    user_id: int,
+    person_id: str,
+    start_ts: int,
+    end_ts: int,
+    transport: str = "",
+) -> list[dict]:
+    backend = get_event_ledger_backend()
+    transport_clause = ""
+    if str(transport or "").strip():
+        transport_clause = (
+            f" AND transport='{backend._escape(str(transport or '').strip().lower())}'"
+        )
+    return backend.query_rows(
+        f"SELECT person_id, session_id, transport, kind, direction, ts, payload "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)} "
+        f"AND person_id='{backend._escape(str(person_id or '').strip())}' "
+        f"AND ts>={int(start_ts)} AND ts<={int(end_ts)}"
+        f"{transport_clause} "
+        "ORDER BY ts ASC"
+    )
+
+
+def get_recent_event_rows(
+    *,
+    minutes: int = 120,
+    service: str = "",
+    user_id: str = "",
+    limit: int = 200,
+) -> list[dict]:
+    backend = get_event_ledger_backend()
+    cutoff_ts = int(time.time() * 1000) - (max(1, int(minutes)) * 60 * 1000)
+    where = [f"ts>={cutoff_ts}"]
+    if service:
+        where.append(f"transport='{backend._escape(str(service).strip().lower())}'")
+    if user_id:
+        where.append(f"user_id={int(user_id)}")
+    rows = backend.query_rows(
+        f"SELECT user_id, session_id, ts, kind, direction, transport, payload "
+        f"FROM {backend.table} "
+        f"WHERE {' AND '.join(where)} "
+        f"ORDER BY ts DESC "
+        f"LIMIT {max(1, min(int(limit), 500))}"
+    )
+    output = []
+    for row in list(rows or []):
+        payload = parse_payload(row.get("payload"))
+        legacy_event_type = str(payload.get("legacy_event_type") or "").strip().lower()
+        output.append(
+            {
+                "id": "",
+                "user_id": int(row.get("user_id") or 0),
+                "session_id": str(row.get("session_id") or ""),
+                "ts": int(row.get("ts") or 0),
+                "event_type": legacy_event_type or str(row.get("kind") or ""),
+                "kind": str(row.get("kind") or ""),
+                "direction": str(row.get("direction") or ""),
+                "origin_transport": str(row.get("transport") or ""),
+                "trace_id": str(payload.get("trace_id") or ""),
+            }
+        )
+    return output
+
+
+def count_behavioral_events(*, user_id: int) -> int:
+    backend = get_event_ledger_backend()
+    rows = backend.query_rows(
+        f"SELECT COUNT(*) AS total_events "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)}"
+    )
+    if not rows:
+        return 0
+    try:
+        return int((rows[0] or {}).get("total_events") or 0)
+    except Exception:
+        return 0
+
+
+def get_trace_ids(*, user_id: int, limit: int = 120) -> list[str]:
+    backend = get_event_ledger_backend()
+    rows = backend.query_rows(
+        f"SELECT payload "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)} "
+        "ORDER BY ts DESC "
+        f"LIMIT {max(1, min(int(limit) * 6, 1000))}"
+    )
+    seen = set()
+    output = []
+    for row in list(rows or []):
+        payload = parse_payload(row.get("payload"))
+        trace_id = str(payload.get("trace_id") or "").strip()
+        if not trace_id or trace_id in seen:
+            continue
+        seen.add(trace_id)
+        output.append(trace_id)
+        if len(output) >= max(1, min(int(limit), 500)):
+            break
+    return output
+
+
+def get_trace_event_rows(*, user_id: int, trace_id: str, limit: int = 500) -> list[dict]:
+    backend = get_event_ledger_backend()
+    rows = backend.query_rows(
+        f"SELECT user_id, session_id, ts, kind, direction, transport, payload "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)} "
+        "ORDER BY ts ASC "
+        f"LIMIT {max(1, min(int(limit) * 8, 5000))}"
+    )
+    output = []
+    target = str(trace_id or "").strip()
+    for row in list(rows or []):
+        payload = parse_payload(row.get("payload"))
+        if str(payload.get("trace_id") or "").strip() != target:
+            continue
+        output.append(
+            {
+                "id": "",
+                "ts": int(row.get("ts") or 0),
+                "event_type": str(
+                    payload.get("legacy_event_type") or row.get("kind") or ""
+                ).strip(),
+                "kind": str(row.get("kind") or "").strip(),
+                "direction": str(row.get("direction") or "").strip(),
+                "session_id": str(row.get("session_id") or "").strip(),
+                "origin_transport": str(row.get("transport") or "").strip(),
+                "origin_message_id": str(payload.get("origin_message_id") or "").strip(),
+                "payload": payload,
+                "trace_id": target,
+            }
+        )
+        if len(output) >= max(1, min(int(limit), 500)):
+            break
+    return output
+
+
+def get_session_event_rows(*, user_id: int, session_id: str, limit: int = 2000) -> list[dict]:
+    backend = get_event_ledger_backend()
+    rows = backend.query_rows(
+        f"SELECT user_id, session_id, ts, kind, direction, transport, actor, payload "
+        f"FROM {backend.table} "
+        f"WHERE user_id={int(user_id)} "
+        f"AND session_id='{backend._escape(str(session_id or '').strip())}' "
+        "ORDER BY ts ASC "
+        f"LIMIT {max(1, min(int(limit), 5000))}"
+    )
+    output = []
+    for row in list(rows or []):
+        payload = parse_payload(row.get("payload"))
+        output.append(
+            {
+                "ts": int(row.get("ts") or 0),
+                "event_type": str(
+                    payload.get("legacy_event_type") or row.get("kind") or ""
+                ).strip(),
+                "kind": str(row.get("kind") or "").strip(),
+                "direction": str(row.get("direction") or "").strip(),
+                "session_id": str(row.get("session_id") or "").strip(),
+                "origin_transport": str(row.get("transport") or "").strip(),
+                "actor_identifier": str(row.get("actor") or "").strip(),
+                "origin_message_id": str(payload.get("origin_message_id") or "").strip(),
+                "payload": payload,
+            }
+        )
+    return output
--- a/core/events/projection.py
+++ b/core/events/projection.py
@@ -0,0 +1,362 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from core.events.manticore import get_session_event_rows
+from core.models import ChatSession, ConversationEvent, Message
+
+
+@dataclass
+class _ProjectedMessage:
+    message_id: str
+    ts: int = 0
+    text: str = ""
+    delivered_ts: int | None = None
+    read_ts: int | None = None
+    reactions: dict[tuple[str, str, str], dict] | None = None
+
+    def __post_init__(self):
+        if self.reactions is None:
+            self.reactions = {}
+
+
+def _safe_int(value, default=0) -> int:
+    try:
+        return int(value)
+    except Exception:
+        return int(default)
+
+
+def _reaction_key(row: dict) -> tuple[str, str, str]:
+    item = dict(row or {})
+    return (
+        str(item.get("source_service") or "").strip().lower(),
+        str(item.get("actor") or "").strip(),
+        str(item.get("emoji") or "").strip(),
+    )
+
+
+def _normalize_reactions(rows: list[dict] | None) -> list[dict]:
+    merged = {}
+    for row in list(rows or []):
+        item = dict(row or {})
+        key = _reaction_key(item)
+        if not any(key):
+            continue
+        merged[key] = {
+            "source_service": key[0],
+            "actor": key[1],
+            "emoji": key[2],
+            "removed": bool(item.get("removed")),
+        }
+    return sorted(
+        merged.values(),
+        key=lambda entry: (
+            str(entry.get("source_service") or ""),
+            str(entry.get("actor") or ""),
+            str(entry.get("emoji") or ""),
+            bool(entry.get("removed")),
+        ),
+    )
+
+
+def _event_rows_for_session(session: ChatSession):
+    try:
+        rows = get_session_event_rows(
+            user_id=int(session.user_id),
+            session_id=str(session.id),
+            limit=2000,
+        )
+    except Exception:
+        rows = []
+    if rows:
+        return rows, "manticore"
+    return (
+        list(
+            ConversationEvent.objects.filter(
+                user=session.user,
+                session=session,
+            ).order_by("ts", "created_at")
+        ),
+        "django",
+    )
+
+
+def project_session_from_events(session: ChatSession) -> list[dict]:
+    rows, _source = _event_rows_for_session(session)
+
+    projected: dict[str, _ProjectedMessage] = {}
+    order: list[str] = []
+
+    for event in rows:
+        is_dict = isinstance(event, dict)
+        payload = dict(
+            (event.get("payload") if is_dict else getattr(event, "payload", {})) or {}
+        )
+        event_type = str(
+            (event.get("event_type") if is_dict else getattr(event, "event_type", ""))
+            or ""
+        ).strip().lower()
+        message_id = str(
+            payload.get("message_id") or payload.get("target_message_id") or ""
+        ).strip()
+
+        if event_type == "message_created":
+            message_id = str(
+                payload.get("message_id")
+                or (
+                    event.get("origin_message_id")
+                    if is_dict
+                    else getattr(event, "origin_message_id", "")
+                )
+                or ""
+            ).strip()
+            if not message_id:
+                continue
+            state = projected.get(message_id)
+            if state is None:
+                state = _ProjectedMessage(message_id=message_id)
+                projected[message_id] = state
+                order.append(message_id)
+            state.ts = _safe_int(
+                payload.get("message_ts"),
+                _safe_int(event.get("ts") if is_dict else getattr(event, "ts", 0)),
+            )
+            state.text = str(payload.get("text") or state.text or "")
+            delivered_default = _safe_int(
+                payload.get("delivered_ts"),
+                _safe_int(event.get("ts") if is_dict else getattr(event, "ts", 0)),
+            )
+            if state.delivered_ts is None:
+                state.delivered_ts = delivered_default or None
+            continue
+
+        if not message_id or message_id not in projected:
+            continue
+        state = projected[message_id]
+
+        if event_type == "read_receipt":
+            read_ts = _safe_int(
+                payload.get("read_ts"),
+                _safe_int(event.get("ts") if is_dict else getattr(event, "ts", 0)),
+            )
+            if read_ts > 0:
+                if state.read_ts is None:
+                    state.read_ts = read_ts
+                else:
+                    state.read_ts = max(int(state.read_ts or 0), read_ts)
+            if state.delivered_ts is None and read_ts > 0:
+                state.delivered_ts = read_ts
+            continue
+
+        if event_type in {"reaction_added", "reaction_removed"}:
+            source_service = (
+                str(
+                    payload.get("source_service")
+                    or (
+                        event.get("origin_transport")
+                        if is_dict
+                        else getattr(event, "origin_transport", "")
+                    )
+                    or ""
+                )
+                .strip()
+                .lower()
+            )
+            actor = str(
+                payload.get("actor")
+                or (
+                    event.get("actor_identifier")
+                    if is_dict
+                    else getattr(event, "actor_identifier", "")
+                )
+                or ""
+            ).strip()
+            emoji = str(payload.get("emoji") or "").strip()
+            if not source_service and not actor and not emoji:
+                continue
+            key = (source_service, actor, emoji)
+            state.reactions[key] = {
+                "source_service": source_service,
+                "actor": actor,
+                "emoji": emoji,
+                "removed": bool(
+                    event_type == "reaction_removed" or payload.get("remove")
+                ),
+            }
+
+    output = []
+    for message_id in order:
+        state = projected.get(message_id)
+        if state is None:
+            continue
+        output.append(
+            {
+                "message_id": str(state.message_id),
+                "ts": int(state.ts or 0),
+                "text": str(state.text or ""),
+                "delivered_ts": (
+                    int(state.delivered_ts) if state.delivered_ts is not None else None
+                ),
+                "read_ts": int(state.read_ts) if state.read_ts is not None else None,
+                "reactions": _normalize_reactions(
+                    list((state.reactions or {}).values())
+                ),
+            }
+        )
+    return output
+
+
+def shadow_compare_session(session: ChatSession, detail_limit: int = 50) -> dict:
+    projected_rows = project_session_from_events(session)
+    projected_by_id = {str(row.get("message_id") or ""): row for row in projected_rows}
+
+    db_rows = list(
+        Message.objects.filter(user=session.user, session=session)
+        .order_by("ts", "id")
+        .values(
+            "id",
+            "ts",
+            "text",
+            "delivered_ts",
+            "read_ts",
+            "receipt_payload",
+        )
+    )
+    db_by_id = {str(row.get("id")): dict(row) for row in db_rows}
+
+    counters = {
+        "missing_in_projection": 0,
+        "missing_in_db": 0,
+        "text_mismatch": 0,
+        "ts_mismatch": 0,
+        "delivered_ts_mismatch": 0,
+        "read_ts_mismatch": 0,
+        "reactions_mismatch": 0,
+    }
+    details = []
+    cause_counts = {
+        "missing_event_write": 0,
+        "ambiguous_reaction_target": 0,
+        "payload_normalization_gap": 0,
+    }
+    cause_samples = {key: [] for key in cause_counts.keys()}
+    cause_sample_limit = min(5, max(0, int(detail_limit)))
+
+    def _record_detail(
+        message_id: str, issue: str, cause: str, extra: dict | None = None
+    ):
+        if cause in cause_counts:
+            cause_counts[cause] += 1
+        row = {"message_id": message_id, "issue": issue, "cause": cause}
+        if extra:
+            row.update(dict(extra))
+        if len(details) < max(0, int(detail_limit)):
+            details.append(row)
+        if cause in cause_samples and len(cause_samples[cause]) < cause_sample_limit:
+            cause_samples[cause].append(row)
+
+    for message_id, db_row in db_by_id.items():
+        projected = projected_by_id.get(message_id)
+        if projected is None:
+            counters["missing_in_projection"] += 1
+            _record_detail(message_id, "missing_in_projection", "missing_event_write")
+            continue
+
+        db_text = str(db_row.get("text") or "")
+        projected_text = str(projected.get("text") or "")
+        if db_text != projected_text:
+            counters["text_mismatch"] += 1
+            _record_detail(
+                message_id,
+                "text_mismatch",
+                "payload_normalization_gap",
+                {"db": db_text, "projected": projected_text},
+            )
+
+        db_ts = _safe_int(db_row.get("ts"), 0)
+        projected_ts = _safe_int(projected.get("ts"), 0)
+        if db_ts != projected_ts:
+            counters["ts_mismatch"] += 1
+            _record_detail(
+                message_id,
+                "ts_mismatch",
+                "payload_normalization_gap",
+                {"db": db_ts, "projected": projected_ts},
+            )
+
+        db_delivered_ts = db_row.get("delivered_ts")
+        projected_delivered_ts = projected.get("delivered_ts")
+        if (db_delivered_ts is None) != (projected_delivered_ts is None) or (
+            db_delivered_ts is not None
+            and projected_delivered_ts is not None
+            and int(db_delivered_ts) != int(projected_delivered_ts)
+        ):
+            counters["delivered_ts_mismatch"] += 1
+            _record_detail(
+                message_id,
+                "delivered_ts_mismatch",
+                "payload_normalization_gap",
+                {
+                    "db": db_delivered_ts,
+                    "projected": projected_delivered_ts,
+                },
+            )
+
+        db_read_ts = db_row.get("read_ts")
+        projected_read_ts = projected.get("read_ts")
+        if (db_read_ts is None) != (projected_read_ts is None) or (
+            db_read_ts is not None
+            and projected_read_ts is not None
+            and int(db_read_ts) != int(projected_read_ts)
+        ):
+            counters["read_ts_mismatch"] += 1
+            _record_detail(
+                message_id,
+                "read_ts_mismatch",
+                "payload_normalization_gap",
+                {"db": db_read_ts, "projected": projected_read_ts},
+            )
+
+        db_reactions = _normalize_reactions(
+            list((db_row.get("receipt_payload") or {}).get("reactions") or [])
+        )
+        projected_reactions = _normalize_reactions(
+            list(projected.get("reactions") or [])
+        )
+        if db_reactions != projected_reactions:
+            counters["reactions_mismatch"] += 1
+            cause = "payload_normalization_gap"
+            strategy = str(
+                (
+                    (db_row.get("receipt_payload") or {}).get(
+                        "reaction_last_match_strategy"
+                    )
+                    or ""
+                )
+            ).strip()
+            if strategy == "nearest_ts_window":
+                cause = "ambiguous_reaction_target"
+            _record_detail(
+                message_id,
+                "reactions_mismatch",
+                cause,
+                {"db": db_reactions, "projected": projected_reactions},
+            )
+
+    for message_id in projected_by_id.keys():
+        if message_id not in db_by_id:
+            counters["missing_in_db"] += 1
+            _record_detail(message_id, "missing_in_db", "payload_normalization_gap")
+
+    mismatch_total = int(sum(int(value or 0) for value in counters.values()))
+    return {
+        "session_id": str(session.id),
+        "db_message_count": len(db_rows),
+        "projected_message_count": len(projected_rows),
+        "mismatch_total": mismatch_total,
+        "counters": counters,
+        "cause_counts": cause_counts,
+        "cause_samples": cause_samples,
+        "details": details,
+    }
--- a/core/events/shadow.py
+++ b/core/events/shadow.py
@@ -0,0 +1,148 @@
+from __future__ import annotations
+
+from django.db.models import Count, Max, Q
+
+from core.models import ConversationEvent, Person, User
+
+
+def _kind_from_event_type(event_type: str) -> str:
+    normalized = str(event_type or "").strip().lower()
+    return {
+        "message_created": "message_sent",
+        "delivery_receipt": "message_delivered",
+        "read_receipt": "message_read",
+        "typing_started": "composing_started",
+        "typing_stopped": "composing_stopped",
+        "composing_abandoned": "composing_abandoned",
+        "presence_available": "presence_available",
+        "presence_unavailable": "presence_unavailable",
+    }.get(normalized, normalized)
+
+
+def get_shadow_behavioral_availability_stats(*, user: User) -> list[dict]:
+    person_map = {
+        str(row["id"]): str(row["name"] or "")
+        for row in Person.objects.filter(user=user).values("id", "name")
+    }
+    rows = (
+        ConversationEvent.objects.filter(
+            user=user,
+            session__identifier__person__isnull=False,
+        )
+        .values("session__identifier__person_id", "origin_transport")
+        .annotate(
+            total_events=Count("id"),
+            presence_events=Count(
+                "id",
+                filter=Q(event_type__in=["presence_available", "presence_unavailable"]),
+            ),
+            read_events=Count("id", filter=Q(event_type="read_receipt")),
+            typing_events=Count(
+                "id",
+                filter=Q(
+                    event_type__in=["typing_started", "typing_stopped"]
+                ),
+            ),
+            message_events=Count("id", filter=Q(event_type="message_created")),
+            abandoned_events=Count("id", filter=Q(event_type="composing_abandoned")),
+            last_event_ts=Max("ts"),
+        )
+        .order_by("-total_events", "session__identifier__person_id", "origin_transport")
+    )
+    output = []
+    for row in rows:
+        person_id = str(row.get("session__identifier__person_id") or "").strip()
+        output.append(
+            {
+                "person_id": person_id,
+                "person_name": person_map.get(person_id, person_id or "-"),
+                "service": str(row.get("origin_transport") or "").strip().lower(),
+                "total_events": int(row.get("total_events") or 0),
+                "presence_events": int(row.get("presence_events") or 0),
+                "read_events": int(row.get("read_events") or 0),
+                "typing_events": int(row.get("typing_events") or 0),
+                "message_events": int(row.get("message_events") or 0),
+                "abandoned_events": int(row.get("abandoned_events") or 0),
+                "last_event_ts": int(row.get("last_event_ts") or 0),
+            }
+        )
+    return output
+
+
+def get_shadow_behavioral_latest_states(
+    *, user: User, person_ids: list[str], transport: str = ""
+) -> list[dict]:
+    queryset = ConversationEvent.objects.filter(
+        user=user,
+        session__identifier__person_id__in=[str(value) for value in person_ids],
+        event_type__in=[
+            "message_created",
+            "delivery_receipt",
+            "read_receipt",
+            "typing_started",
+            "typing_stopped",
+            "composing_abandoned",
+            "presence_available",
+            "presence_unavailable",
+        ],
+    ).select_related("session__identifier")
+    if transport:
+        queryset = queryset.filter(origin_transport=str(transport).strip().lower())
+    rows = []
+    seen = set()
+    for row in queryset.order_by(
+        "session__identifier__person_id", "-ts", "-created_at"
+    )[:500]:
+        person_id = str(getattr(row.session.identifier, "person_id", "") or "").strip()
+        if not person_id or person_id in seen:
+            continue
+        seen.add(person_id)
+        rows.append(
+            {
+                "person_id": person_id,
+                "transport": str(row.origin_transport or "").strip().lower(),
+                "kind": _kind_from_event_type(row.event_type),
+                "ts": int(row.ts or 0),
+            }
+        )
+    return rows
+
+
+def get_shadow_behavioral_events_for_range(
+    *,
+    user: User,
+    person_id: str,
+    start_ts: int,
+    end_ts: int,
+    transport: str = "",
+) -> list[dict]:
+    queryset = ConversationEvent.objects.filter(
+        user=user,
+        session__identifier__person_id=str(person_id or "").strip(),
+        ts__gte=int(start_ts),
+        ts__lte=int(end_ts),
+        event_type__in=[
+            "message_created",
+            "delivery_receipt",
+            "read_receipt",
+            "typing_started",
+            "typing_stopped",
+            "composing_abandoned",
+            "presence_available",
+            "presence_unavailable",
+        ],
+    ).order_by("ts", "created_at")
+    if transport:
+        queryset = queryset.filter(origin_transport=str(transport).strip().lower())
+    return [
+        {
+            "person_id": str(person_id or "").strip(),
+            "session_id": str(row.session_id or ""),
+            "transport": str(row.origin_transport or "").strip().lower(),
+            "kind": _kind_from_event_type(row.event_type),
+            "direction": str(row.direction or "").strip().lower(),
+            "ts": int(row.ts or 0),
+            "payload": dict(row.payload or {}),
+        }
+        for row in queryset[:1000]
+    ]
--- a/core/forms.py
+++ b/core/forms.py
@@ -1,6 +1,7 @@
 from django import forms
 from django.contrib.auth.forms import UserCreationForm
 from django.forms import ModelForm
+
 from mixins.restrictions import RestrictedFormMixin

 from .models import (
--- a/core/gateway/init.py
+++ b/core/gateway/init.py
@@ -0,0 +1 @@
+"""Gateway command routing utilities."""
--- a/core/gateway/builtin.py
+++ b/core/gateway/builtin.py
@@ -0,0 +1,417 @@
+from __future__ import annotations
+
+import re
+
+from asgiref.sync import sync_to_async
+
+from core.gateway.commands import (
+    GatewayCommandContext,
+    GatewayCommandRoute,
+    dispatch_gateway_command,
+)
+from core.models import (
+    CodexPermissionRequest,
+    CodexRun,
+    DerivedTask,
+    ExternalSyncEvent,
+    Person,
+    TaskProject,
+    User,
+)
+from core.tasks.engine import create_task_record_and_sync, mark_task_completed_and_sync
+
+APPROVAL_PROVIDER_COMMANDS = {
+    ".claude": "claude",
+    ".codex": "codex_cli",
+}
+APPROVAL_EVENT_PREFIX = "codex_approval"
+ACTION_TO_STATUS = {"approve": "approved", "reject": "denied"}
+TASK_COMMAND_MATCH_RE = re.compile(r"^\s*(?:\.tasks\b|\.l\b|\.list\b)", re.IGNORECASE)
+
+
+def gateway_help_lines() -> list[str]:
+    return [
+        "Gateway commands:",
+        "  .contacts — list contacts",
+        "  .whoami — show current user",
+        "  .help — show this help",
+        "Approval commands:",
+        "  .approval list-pending [all] — list pending approval requests",
+        "  .approval approve <key> — approve a request",
+        "  .approval reject <key> — reject a request",
+        "  .approval status <key> — check request status",
+        "Task commands:",
+        "  .l — shortcut for open task list",
+        "  .tasks list [status] [limit] — list tasks",
+        "  .tasks add <project> :: <title> — create task in project",
+        "  .tasks show #<ref> — show task details",
+        "  .tasks complete #<ref> — mark task complete",
+        "  .tasks undo #<ref> — remove task",
+    ]
+
+
+def _resolve_request_provider(request):
+    event = getattr(request, "external_sync_event", None)
+    if event is None:
+        return ""
+    return str(getattr(event, "provider", "") or "").strip()
+
+
+async def _apply_approval_decision(request, decision):
+    status = ACTION_TO_STATUS.get(decision, decision)
+    request.status = status
+    await sync_to_async(request.save)(update_fields=["status"])
+    run = None
+    if request.codex_run_id:
+        run = await sync_to_async(CodexRun.objects.get)(pk=request.codex_run_id)
+        run.status = "approved_waiting_resume" if status == "approved" else status
+        await sync_to_async(run.save)(update_fields=["status"])
+    if request.external_sync_event_id:
+        evt = await sync_to_async(ExternalSyncEvent.objects.get)(
+            pk=request.external_sync_event_id
+        )
+        evt.status = "ok"
+        await sync_to_async(evt.save)(update_fields=["status"])
+    user = await sync_to_async(User.objects.get)(pk=request.user_id)
+    task = None
+    if run is not None and run.task_id:
+        task = await sync_to_async(DerivedTask.objects.get)(pk=run.task_id)
+    ikey = f"{APPROVAL_EVENT_PREFIX}:{request.approval_key}:{status}"
+    await sync_to_async(ExternalSyncEvent.objects.get_or_create)(
+        idempotency_key=ikey,
+        defaults={
+            "user": user,
+            "task": task,
+            "provider": "codex_cli",
+            "status": "pending",
+            "payload": {},
+            "error": "",
+        },
+    )
+
+
+async def _approval_list_pending(user, scope, emit):
+    _ = scope
+    requests = await sync_to_async(list)(
+        CodexPermissionRequest.objects.filter(user=user, status="pending").order_by(
+            "-requested_at"
+        )[:20]
+    )
+    emit(f"pending={len(requests)}")
+    for req in requests:
+        emit(f"  {req.approval_key}: {req.summary}")
+
+
+async def _approval_status(user, approval_key, emit):
+    try:
+        req = await sync_to_async(CodexPermissionRequest.objects.get)(
+            user=user, approval_key=approval_key
+        )
+        emit(f"status={req.status} key={req.approval_key}")
+    except CodexPermissionRequest.DoesNotExist:
+        emit(f"approval_key_not_found:{approval_key}")
+
+
+async def handle_approval_command(user, body, emit):
+    command = str(body or "").strip()
+    for prefix, expected_provider in APPROVAL_PROVIDER_COMMANDS.items():
+        if command.startswith(prefix + " ") or command == prefix:
+            sub = command[len(prefix) :].strip()
+            parts = sub.split()
+            if len(parts) >= 2 and parts[0] in ("approve", "reject"):
+                action, approval_key = parts[0], parts[1]
+                try:
+                    req = await sync_to_async(
+                        CodexPermissionRequest.objects.select_related(
+                            "external_sync_event"
+                        ).get
+                    )(user=user, approval_key=approval_key)
+                except CodexPermissionRequest.DoesNotExist:
+                    emit(f"approval_key_not_found:{approval_key}")
+                    return True
+                provider = _resolve_request_provider(req)
+                if not provider.startswith(expected_provider):
+                    emit(
+                        f"approval_key_not_for_provider:{approval_key} provider={provider}"
+                    )
+                    return True
+                await _apply_approval_decision(req, action)
+                emit(f"{action}d: {approval_key}")
+                return True
+            emit(f"usage: {prefix} approve|reject <key>")
+            return True
+
+    if not command.startswith(".approval"):
+        return False
+
+    rest = command[len(".approval") :].strip()
+
+    if rest.split() and rest.split()[0] in ("approve", "reject"):
+        parts = rest.split()
+        action = parts[0]
+        approval_key = parts[1] if len(parts) > 1 else ""
+        if not approval_key:
+            emit("usage: .approval approve|reject <key>")
+            return True
+        try:
+            req = await sync_to_async(
+                CodexPermissionRequest.objects.select_related("external_sync_event").get
+            )(user=user, approval_key=approval_key)
+        except CodexPermissionRequest.DoesNotExist:
+            emit(f"approval_key_not_found:{approval_key}")
+            return True
+        await _apply_approval_decision(req, action)
+        emit(f"{action}d: {approval_key}")
+        return True
+
+    if rest.startswith("list-pending"):
+        scope = rest[len("list-pending") :].strip() or "mine"
+        await _approval_list_pending(user, scope, emit)
+        return True
+
+    if rest.startswith("status "):
+        approval_key = rest[len("status ") :].strip()
+        await _approval_status(user, approval_key, emit)
+        return True
+
+    emit(
+        "approval: .approval approve|reject <key> | "
+        ".approval list-pending [all] | "
+        ".approval status <key>"
+    )
+    return True
+
+
+def _parse_task_create(rest: str) -> tuple[str, str]:
+    text = str(rest or "").strip()
+    if not text.lower().startswith("add "):
+        return "", ""
+    payload = text[4:].strip()
+    if "::" in payload:
+        project_name, title = payload.split("::", 1)
+        return str(project_name or "").strip(), str(title or "").strip()
+    return "", ""
+
+
+async def handle_tasks_command(
+    user,
+    body,
+    emit,
+    *,
+    service: str = "",
+    channel_identifier: str = "",
+    sender_identifier: str = "",
+):
+    command = str(body or "").strip()
+    lower_command = command.lower()
+    if not TASK_COMMAND_MATCH_RE.match(command):
+        return False
+    if lower_command.startswith(".tasks"):
+        rest = command[len(".tasks") :].strip()
+    elif lower_command.startswith(".list") or lower_command.startswith(".l"):
+        rest = "list"
+    else:
+        rest = "list " + command[2:].strip() if len(command) > 2 else "list"
+
+    if rest.startswith("list"):
+        parts = rest.split()
+        status_filter = parts[1] if len(parts) > 1 else "open"
+        limit = int(parts[2]) if len(parts) > 2 and parts[2].isdigit() else 10
+        tasks = await sync_to_async(list)(
+            DerivedTask.objects.filter(
+                user=user, status_snapshot=status_filter
+            ).order_by("-id")[:limit]
+        )
+        if not tasks:
+            emit(f"no {status_filter} tasks")
+        else:
+            for task in tasks:
+                emit(f"#{task.reference_code} [{task.status_snapshot}] {task.title}")
+        return True
+
+    project_name, title = _parse_task_create(rest)
+    if project_name or rest.startswith("add "):
+        if not project_name or not title:
+            emit("usage: .tasks add <project> :: <title>")
+            return True
+        project = await sync_to_async(
+            lambda: TaskProject.objects.filter(user=user, name__iexact=project_name)
+            .order_by("name")
+            .first()
+        )()
+        if project is None:
+            emit(f"project_not_found:{project_name}")
+            return True
+        task, _event = await create_task_record_and_sync(
+            user=user,
+            project=project,
+            title=title,
+            source_service=str(service or "web").strip().lower() or "web",
+            source_channel=str(channel_identifier or "").strip(),
+            actor_identifier=str(sender_identifier or "").strip(),
+            immutable_payload={
+                "origin": "gateway.tasks.add",
+                "channel_service": str(service or "").strip().lower(),
+                "channel_identifier": str(channel_identifier or "").strip(),
+            },
+            event_payload={
+                "command": ".tasks add",
+                "via": "gateway_builtin",
+            },
+        )
+        emit(f"created #{task.reference_code} [{project.name}] {task.title}")
+        return True
+
+    if rest.startswith("show "):
+        ref = rest[len("show ") :].strip().lstrip("#")
+        try:
+            task = await sync_to_async(DerivedTask.objects.get)(
+                user=user, reference_code=ref
+            )
+            emit(f"#{task.reference_code} {task.title}")
+            emit(f"status: {task.status_snapshot}")
+        except DerivedTask.DoesNotExist:
+            emit(f"task_not_found:#{ref}")
+        return True
+
+    if rest.startswith("complete "):
+        ref = rest[len("complete ") :].strip().lstrip("#")
+        try:
+            task = await sync_to_async(DerivedTask.objects.select_related("project").get)(
+                user=user, reference_code=ref
+            )
+            await mark_task_completed_and_sync(
+                task=task,
+                actor_identifier=str(sender_identifier or "").strip(),
+                payload={
+                    "marker": ref,
+                    "command": ".tasks complete",
+                    "via": "gateway_builtin",
+                },
+            )
+            emit(f"completed #{ref}")
+        except DerivedTask.DoesNotExist:
+            emit(f"task_not_found:#{ref}")
+        return True
+
+    if rest.startswith("undo "):
+        ref = rest[len("undo ") :].strip().lstrip("#")
+        try:
+            task = await sync_to_async(DerivedTask.objects.get)(
+                user=user, reference_code=ref
+            )
+            await sync_to_async(task.delete)()
+            emit(f"removed #{ref}")
+        except DerivedTask.DoesNotExist:
+            emit(f"task_not_found:#{ref}")
+        return True
+
+    emit(
+        "tasks: .l | .tasks list [status] [limit] | "
+        ".tasks add <project> :: <title> | "
+        ".tasks show #<ref> | "
+        ".tasks complete #<ref> | "
+        ".tasks undo #<ref>"
+    )
+    return True
+
+
+async def dispatch_builtin_gateway_command(
+    *,
+    user,
+    command_text: str,
+    service: str,
+    channel_identifier: str,
+    sender_identifier: str,
+    source_message,
+    message_meta: dict,
+    payload: dict,
+    emit,
+) -> bool:
+    text = str(command_text or "").strip()
+
+    async def _contacts_handler(_ctx, out):
+        persons = await sync_to_async(list)(Person.objects.filter(user=user).order_by("name"))
+        if not persons:
+            out("No contacts found.")
+            return True
+        out("Contacts: " + ", ".join([p.name for p in persons]))
+        return True
+
+    async def _help_handler(_ctx, out):
+        for line in gateway_help_lines():
+            out(line)
+        return True
+
+    async def _whoami_handler(_ctx, out):
+        out(str(user.__dict__))
+        return True
+
+    async def _approval_handler(_ctx, out):
+        return await handle_approval_command(user, text, out)
+
+    async def _tasks_handler(_ctx, out):
+        return await handle_tasks_command(
+            user,
+            text,
+            out,
+            service=service,
+            channel_identifier=channel_identifier,
+            sender_identifier=sender_identifier,
+        )
+
+    routes = [
+        GatewayCommandRoute(
+            name="contacts",
+            scope_key="gateway.contacts",
+            matcher=lambda value: str(value or "").strip().lower() == ".contacts",
+            handler=_contacts_handler,
+        ),
+        GatewayCommandRoute(
+            name="help",
+            scope_key="gateway.help",
+            matcher=lambda value: str(value or "").strip().lower() == ".help",
+            handler=_help_handler,
+        ),
+        GatewayCommandRoute(
+            name="whoami",
+            scope_key="gateway.whoami",
+            matcher=lambda value: str(value or "").strip().lower() == ".whoami",
+            handler=_whoami_handler,
+        ),
+        GatewayCommandRoute(
+            name="approval",
+            scope_key="gateway.approval",
+            matcher=lambda value: str(value or "").strip().lower().startswith(".approval")
+            or any(
+                str(value or "").strip().lower().startswith(prefix + " ")
+                or str(value or "").strip().lower() == prefix
+                for prefix in APPROVAL_PROVIDER_COMMANDS
+            ),
+            handler=_approval_handler,
+        ),
+        GatewayCommandRoute(
+            name="tasks",
+            scope_key="gateway.tasks",
+            matcher=lambda value: bool(TASK_COMMAND_MATCH_RE.match(str(value or ""))),
+            handler=_tasks_handler,
+        ),
+    ]
+
+    handled = await dispatch_gateway_command(
+        context=GatewayCommandContext(
+            user=user,
+            source_message=source_message,
+            service=str(service or "xmpp"),
+            channel_identifier=str(channel_identifier or ""),
+            sender_identifier=str(sender_identifier or ""),
+            message_text=text,
+            message_meta=dict(message_meta or {}),
+            payload=dict(payload or {}),
+        ),
+        routes=routes,
+        emit=emit,
+    )
+    if not handled and text.startswith("."):
+        emit("No such command")
+    return handled
--- a/core/gateway/commands.py
+++ b/core/gateway/commands.py
@@ -0,0 +1,137 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Awaitable, Callable
+
+from asgiref.sync import sync_to_async
+
+from core.models import GatewayCommandEvent
+from core.security.command_policy import CommandSecurityContext, evaluate_command_policy
+
+GatewayEmit = Callable[[str], None]
+GatewayHandler = Callable[["GatewayCommandContext", GatewayEmit], Awaitable[bool]]
+GatewayMatcher = Callable[[str], bool]
+
+
+@dataclass(slots=True)
+class GatewayCommandContext:
+    user: object
+    source_message: object
+    service: str
+    channel_identifier: str
+    sender_identifier: str
+    message_text: str
+    message_meta: dict
+    payload: dict
+
+
+@dataclass(slots=True)
+class GatewayCommandRoute:
+    name: str
+    scope_key: str
+    matcher: GatewayMatcher
+    handler: GatewayHandler
+
+
+def _first_token(text: str) -> str:
+    body = str(text or "").strip()
+    if not body:
+        return ""
+    return str(body.split()[0] or "").strip().lower()
+
+
+def _derive_unknown_scope(text: str) -> str:
+    token = _first_token(text).lstrip(".")
+    if not token:
+        token = "message"
+    return f"gateway.{token}"
+
+
+async def dispatch_gateway_command(
+    *,
+    context: GatewayCommandContext,
+    routes: list[GatewayCommandRoute],
+    emit: GatewayEmit,
+) -> bool:
+    text = str(context.message_text or "").strip()
+    if not text:
+        return False
+
+    route = next((row for row in routes if row.matcher(text)), None)
+    scope_key = route.scope_key if route is not None else _derive_unknown_scope(text)
+    command_name = route.name if route is not None else _first_token(text).lstrip(".")
+
+    event = await sync_to_async(GatewayCommandEvent.objects.create)(
+        user=context.user,
+        source_message=context.source_message,
+        service=str(context.service or "").strip().lower() or "xmpp",
+        channel_identifier=str(context.channel_identifier or "").strip(),
+        sender_identifier=str(context.sender_identifier or "").strip(),
+        scope_key=scope_key,
+        command_name=command_name,
+        command_text=text,
+        status="pending",
+        request_meta={
+            "payload": dict(context.payload or {}),
+            "message_meta": dict(context.message_meta or {}),
+        },
+    )
+
+    if route is None:
+        event.status = "ignored"
+        event.error = "unmatched_gateway_command"
+        await sync_to_async(event.save)(update_fields=["status", "error", "updated_at"])
+        return False
+
+    decision = await sync_to_async(evaluate_command_policy)(
+        user=context.user,
+        scope_key=scope_key,
+        context=CommandSecurityContext(
+            service=context.service,
+            channel_identifier=context.channel_identifier,
+            message_meta=dict(context.message_meta or {}),
+            payload=dict(context.payload or {}),
+        ),
+    )
+    if not decision.allowed:
+        message = (
+            f"blocked by policy: {decision.code}"
+            if not decision.reason
+            else f"blocked by policy: {decision.reason}"
+        )
+        emit(message)
+        event.status = "blocked"
+        event.error = f"{decision.code}:{decision.reason}"
+        event.response_meta = {
+            "policy_code": decision.code,
+            "policy_reason": decision.reason,
+        }
+        await sync_to_async(event.save)(
+            update_fields=["status", "error", "response_meta", "updated_at"]
+        )
+        return True
+
+    responses: list[str] = []
+
+    def _captured_emit(value: str) -> None:
+        row = str(value or "")
+        responses.append(row)
+        emit(row)
+
+    try:
+        handled = await route.handler(context, _captured_emit)
+    except Exception as exc:
+        event.status = "failed"
+        event.error = f"handler_exception:{exc}"
+        event.response_meta = {"responses": responses}
+        await sync_to_async(event.save)(
+            update_fields=["status", "error", "response_meta", "updated_at"]
+        )
+        return True
+
+    event.status = "ok" if handled else "ignored"
+    event.response_meta = {"responses": responses}
+    await sync_to_async(event.save)(
+        update_fields=["status", "response_meta", "updated_at"]
+    )
+    return bool(handled)
--- a/core/management/commands/backfill_contact_availability.py
+++ b/core/management/commands/backfill_contact_availability.py
@@ -0,0 +1,128 @@
+from __future__ import annotations
+
+from typing import Iterable
+
+from django.core.management.base import BaseCommand
+
+from core.events.ledger import append_event_sync
+from core.models import Message
+from core.presence.inference import now_ms
+
+
+class Command(BaseCommand):
+    help = (
+        "Backfill behavioral event ledger rows from historical message and "
+        "read-receipt activity."
+    )
+
+    def add_arguments(self, parser):
+        parser.add_argument("--days", type=int, default=30)
+        parser.add_argument("--limit", type=int, default=5000)
+        parser.add_argument("--service", default="")
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--dry-run", action="store_true", default=False)
+
+    def _iter_messages(
+        self, *, days: int, limit: int, service: str, user_id: str
+    ) -> Iterable[Message]:
+        cutoff_ts = now_ms() - (max(1, int(days)) * 24 * 60 * 60 * 1000)
+        qs = Message.objects.filter(ts__gte=cutoff_ts).select_related(
+            "user", "session", "session__identifier", "session__identifier__person"
+        )
+        if service:
+            qs = qs.filter(source_service=str(service).strip().lower())
+        if user_id:
+            qs = qs.filter(user_id=str(user_id).strip())
+        return qs.order_by("ts")[: max(1, int(limit))]
+
+    def handle(self, *args, **options):
+        days = max(1, int(options.get("days") or 30))
+        limit = max(1, int(options.get("limit") or 5000))
+        service_filter = str(options.get("service") or "").strip().lower()
+        user_filter = str(options.get("user_id") or "").strip()
+        dry_run = bool(options.get("dry_run"))
+
+        indexed = 0
+        scanned = 0
+
+        for msg in self._iter_messages(
+            days=days, limit=limit, service=service_filter, user_id=user_filter
+        ):
+            scanned += 1
+            session = getattr(msg, "session", None)
+            identifier = getattr(session, "identifier", None)
+            person = getattr(identifier, "person", None)
+            user = getattr(msg, "user", None)
+            if not session or not identifier or not person or not user:
+                continue
+
+            service = (
+                str(getattr(msg, "source_service", "") or identifier.service or "")
+                .strip()
+                .lower()
+            )
+            if not service:
+                continue
+
+            author = str(getattr(msg, "custom_author", "") or "").strip().upper()
+            outgoing = author in {"USER", "BOT"}
+            message_id = str(
+                getattr(msg, "source_message_id", "") or f"django-message-{msg.id}"
+            ).strip()
+
+            if not dry_run:
+                append_event_sync(
+                    user=user,
+                    session=session,
+                    ts=int(getattr(msg, "ts", 0) or 0),
+                    event_type="message_created",
+                    direction="out" if outgoing else "in",
+                    actor_identifier=str(
+                        getattr(msg, "sender_uuid", "") or identifier.identifier or ""
+                    ),
+                    origin_transport=service,
+                    origin_message_id=message_id,
+                    origin_chat_id=str(getattr(msg, "source_chat_id", "") or ""),
+                    payload={
+                        "origin": "backfill_contact_availability",
+                        "message_id": str(msg.id),
+                        "text": str(getattr(msg, "text", "") or ""),
+                        "outgoing": outgoing,
+                    },
+                )
+            indexed += 1
+
+            read_ts = int(getattr(msg, "read_ts", 0) or 0)
+            if read_ts <= 0:
+                continue
+            if not dry_run:
+                append_event_sync(
+                    user=user,
+                    session=session,
+                    ts=read_ts,
+                    event_type="read_receipt",
+                    direction="system",
+                    actor_identifier=str(
+                        getattr(msg, "read_by_identifier", "") or identifier.identifier
+                    ),
+                    origin_transport=service,
+                    origin_message_id=message_id,
+                    origin_chat_id=str(getattr(msg, "source_chat_id", "") or ""),
+                    payload={
+                        "origin": "backfill_contact_availability",
+                        "message_id": str(msg.id),
+                        "message_ts": int(getattr(msg, "ts", 0) or 0),
+                        "read_by": str(
+                            getattr(msg, "read_by_identifier", "") or ""
+                        ).strip(),
+                    },
+                )
+            indexed += 1
+
+        self.stdout.write(
+            self.style.SUCCESS(
+                "backfill_contact_availability complete "
+                f"scanned={scanned} indexed={indexed} dry_run={dry_run} "
+                f"days={days} limit={limit}"
+            )
+        )
--- a/core/management/commands/codex_worker.py
+++ b/core/management/commands/codex_worker.py
@@ -0,0 +1,328 @@
+from __future__ import annotations
+
+import time
+import uuid
+
+from asgiref.sync import async_to_sync
+from django.core.management.base import BaseCommand
+
+from core.clients.transport import send_message_raw
+from core.models import (
+    CodexPermissionRequest,
+    CodexRun,
+    ExternalSyncEvent,
+    TaskProviderConfig,
+)
+from core.tasks.providers import get_provider
+from core.util import logs
+
+log = logs.get_logger("codex_worker")
+
+
+class Command(BaseCommand):
+    help = (
+        "Process queued external sync events for worker-backed providers (codex_cli)."
+    )
+
+    def add_arguments(self, parser):
+        parser.add_argument("--once", action="store_true", default=False)
+        parser.add_argument("--sleep-seconds", type=float, default=2.0)
+        parser.add_argument("--batch-size", type=int, default=20)
+        parser.add_argument("--provider", default="codex_cli")
+
+    def _claim_batch(self, provider: str, batch_size: int) -> list[str]:
+        ids: list[str] = []
+        rows = list(
+            ExternalSyncEvent.objects.filter(
+                provider=provider,
+                status__in=["pending", "retrying"],
+            )
+            .order_by("updated_at")[: max(1, batch_size)]
+            .values_list("id", flat=True)
+        )
+        for row_id in rows:
+            updated = ExternalSyncEvent.objects.filter(
+                id=row_id,
+                provider=provider,
+                status__in=["pending", "retrying"],
+            ).update(status="retrying")
+            if updated:
+                ids.append(str(row_id))
+        return ids
+
+    def _run_event(self, event: ExternalSyncEvent) -> None:
+        provider = get_provider(event.provider)
+        if not bool(getattr(provider, "run_in_worker", False)):
+            return
+
+        cfg = (
+            TaskProviderConfig.objects.filter(
+                user=event.user,
+                provider=event.provider,
+                enabled=True,
+            )
+            .order_by("-updated_at")
+            .first()
+        )
+        if cfg is None:
+            event.status = "failed"
+            event.error = "provider_disabled_or_missing"
+            event.save(update_fields=["status", "error", "updated_at"])
+            provider_payload = dict((event.payload or {}).get("provider_payload") or {})
+            run_id = str(provider_payload.get("codex_run_id") or "").strip()
+            if run_id:
+                CodexRun.objects.filter(id=run_id, user=event.user).update(
+                    status="failed",
+                    error="provider_disabled_or_missing",
+                )
+            return
+
+        payload = dict(event.payload or {})
+        action = str(payload.get("action") or "append_update").strip().lower()
+        provider_payload = dict(payload.get("provider_payload") or payload)
+        run_id = str(
+            provider_payload.get("codex_run_id") or payload.get("codex_run_id") or ""
+        ).strip()
+        codex_run = None
+        if run_id:
+            codex_run = CodexRun.objects.filter(id=run_id, user=event.user).first()
+        if codex_run is None and event.task_id:
+            codex_run = (
+                CodexRun.objects.filter(
+                    user=event.user,
+                    task_id=event.task_id,
+                    status__in=["queued", "running", "approved_waiting_resume"],
+                )
+                .order_by("-updated_at")
+                .first()
+            )
+        if codex_run is not None:
+            codex_run.status = "running"
+            codex_run.error = ""
+            codex_run.save(update_fields=["status", "error", "updated_at"])
+
+        if action == "create":
+            result = provider.create_task(dict(cfg.settings or {}), provider_payload)
+        elif action == "complete":
+            result = provider.mark_complete(dict(cfg.settings or {}), provider_payload)
+        elif action == "link_task":
+            result = provider.link_task(dict(cfg.settings or {}), provider_payload)
+        else:
+            result = provider.append_update(dict(cfg.settings or {}), provider_payload)
+
+        result_payload = dict(result.payload or {})
+        requires_approval = bool(result_payload.get("requires_approval"))
+        if requires_approval:
+            approval_key = str(
+                result_payload.get("approval_key") or uuid.uuid4().hex[:12]
+            ).strip()
+            permission_request = dict(result_payload.get("permission_request") or {})
+            summary = str(
+                result_payload.get("summary") or permission_request.get("summary") or ""
+            ).strip()
+            requested_permissions = permission_request.get("requested_permissions")
+            if not isinstance(requested_permissions, (list, dict)):
+                requested_permissions = permission_request or {}
+            resume_payload = result_payload.get("resume_payload")
+            if not isinstance(resume_payload, dict):
+                resume_payload = {}
+            event.status = "waiting_approval"
+            event.error = ""
+            event.payload = dict(payload, worker_processed=True, result=result_payload)
+            event.save(update_fields=["status", "error", "payload", "updated_at"])
+            if codex_run is not None:
+                codex_run.status = "waiting_approval"
+                codex_run.result_payload = dict(result_payload)
+                codex_run.error = ""
+                codex_run.save(
+                    update_fields=["status", "result_payload", "error", "updated_at"]
+                )
+            CodexPermissionRequest.objects.update_or_create(
+                approval_key=approval_key,
+                defaults={
+                    "user": event.user,
+                    "codex_run": (
+                        codex_run
+                        if codex_run is not None
+                        else CodexRun.objects.create(
+                            user=event.user,
+                            task=event.task,
+                            derived_task_event=event.task_event,
+                            source_service=str(
+                                provider_payload.get("source_service") or ""
+                            ),
+                            source_channel=str(
+                                provider_payload.get("source_channel") or ""
+                            ),
+                            external_chat_id=str(
+                                provider_payload.get("external_chat_id") or ""
+                            ),
+                            status="waiting_approval",
+                            request_payload=dict(payload or {}),
+                            result_payload=dict(result_payload),
+                            error="",
+                        )
+                    ),
+                    "external_sync_event": event,
+                    "summary": summary,
+                    "requested_permissions": (
+                        requested_permissions
+                        if isinstance(requested_permissions, dict)
+                        else {"items": list(requested_permissions or [])}
+                    ),
+                    "resume_payload": dict(resume_payload or {}),
+                    "status": "pending",
+                    "resolved_at": None,
+                    "resolved_by_identifier": "",
+                    "resolution_note": "",
+                },
+            )
+            approver_service = (
+                str((cfg.settings or {}).get("approver_service") or "").strip().lower()
+            )
+            approver_identifier = str(
+                (cfg.settings or {}).get("approver_identifier") or ""
+            ).strip()
+            requested_text = (
+                result_payload.get("permission_request")
+                or result_payload.get("requested_permissions")
+                or {}
+            )
+            if approver_service and approver_identifier:
+                try:
+                    async_to_sync(send_message_raw)(
+                        approver_service,
+                        approver_identifier,
+                        text=(
+                            f"[codex approval] key={approval_key}\\n"
+                            f"summary={summary or 'Codex run requires approval'}\\n"
+                            f"requested={requested_text}\\n"
+                            f"use: .codex approve {approval_key} or .codex deny {approval_key}"
+                        ),
+                        attachments=[],
+                        metadata={"origin_tag": f"codex-approval:{approval_key}"},
+                    )
+                except Exception:
+                    log.exception(
+                        "failed to notify approver channel for approval_key=%s",
+                        approval_key,
+                    )
+            else:
+                source_service = (
+                    str(provider_payload.get("source_service") or "").strip().lower()
+                )
+                source_channel = str(
+                    provider_payload.get("source_channel") or ""
+                ).strip()
+                if source_service and source_channel:
+                    try:
+                        async_to_sync(send_message_raw)(
+                            source_service,
+                            source_channel,
+                            text=(
+                                "[codex approval] approval is pending but no approver channel is configured. "
+                                "Set approver_service and approver_identifier in Codex settings."
+                            ),
+                            attachments=[],
+                            metadata={"origin_tag": "codex-approval-missing-target"},
+                        )
+                    except Exception:
+                        log.exception(
+                            "failed to notify source channel for missing approver target"
+                        )
+            return
+
+        event.status = "ok" if result.ok else "failed"
+        event.error = str(result.error or "")
+        event.payload = dict(
+            payload,
+            worker_processed=True,
+            result=result_payload,
+        )
+        event.save(update_fields=["status", "error", "payload", "updated_at"])
+
+        mode = str(provider_payload.get("mode") or "").strip().lower()
+        approval_key = str(provider_payload.get("approval_key") or "").strip()
+        if mode == "approval_response" and approval_key:
+            req = (
+                CodexPermissionRequest.objects.select_related(
+                    "external_sync_event", "codex_run"
+                )
+                .filter(user=event.user, approval_key=approval_key)
+                .first()
+            )
+            if req and req.external_sync_event_id:
+                if result.ok:
+                    ExternalSyncEvent.objects.filter(
+                        id=req.external_sync_event_id
+                    ).update(
+                        status="ok",
+                        error="",
+                    )
+                elif str(event.error or "").strip() == "approval_denied":
+                    ExternalSyncEvent.objects.filter(
+                        id=req.external_sync_event_id
+                    ).update(
+                        status="failed",
+                        error="approval_denied",
+                    )
+        if codex_run is not None:
+            codex_run.status = "ok" if result.ok else "failed"
+            codex_run.error = str(result.error or "")
+            codex_run.result_payload = result_payload
+            codex_run.save(
+                update_fields=["status", "error", "result_payload", "updated_at"]
+            )
+
+        if (
+            result.ok
+            and result.external_key
+            and event.task_id
+            and not str(event.task.external_key or "").strip()
+        ):
+            event.task.external_key = str(result.external_key)
+            event.task.save(update_fields=["external_key"])
+
+    def handle(self, *args, **options):
+        once = bool(options.get("once"))
+        sleep_seconds = max(0.2, float(options.get("sleep_seconds") or 2.0))
+        batch_size = max(1, int(options.get("batch_size") or 20))
+        provider_name = str(options.get("provider") or "codex_cli").strip().lower()
+
+        log.info(
+            "codex_worker started provider=%s once=%s sleep=%s batch_size=%s",
+            provider_name,
+            once,
+            sleep_seconds,
+            batch_size,
+        )
+
+        while True:
+            claimed_ids = self._claim_batch(provider_name, batch_size)
+            if not claimed_ids:
+                if once:
+                    log.info("codex_worker exiting: no pending events")
+                    return
+                time.sleep(sleep_seconds)
+                continue
+
+            for row_id in claimed_ids:
+                event = (
+                    ExternalSyncEvent.objects.filter(id=row_id)
+                    .select_related("task", "user")
+                    .first()
+                )
+                if event is None:
+                    continue
+                try:
+                    self._run_event(event)
+                except Exception as exc:
+                    log.exception("codex_worker failed processing id=%s", row_id)
+                    ExternalSyncEvent.objects.filter(id=row_id).update(
+                        status="failed",
+                        error=f"worker_exception:{exc}",
+                    )
+
+            if once:
+                log.info("codex_worker processed %s event(s)", len(claimed_ids))
+                return
--- a/core/management/commands/event_ledger_smoke.py
+++ b/core/management/commands/event_ledger_smoke.py
@@ -0,0 +1,117 @@
+from __future__ import annotations
+
+import json
+import time
+
+from django.core.management.base import BaseCommand, CommandError
+
+from core.events.manticore import get_recent_event_rows
+from core.models import ConversationEvent
+
+
+class Command(BaseCommand):
+    help = "Quick non-mutating sanity check for recent canonical event writes."
+
+    def _recent_rows(self, *, minutes: int, service: str, user_id: str, limit: int):
+        cutoff_ts = int(time.time() * 1000) - (minutes * 60 * 1000)
+        queryset = ConversationEvent.objects.filter(ts__gte=cutoff_ts).order_by("-ts")
+        if service:
+            queryset = queryset.filter(origin_transport=service)
+        if user_id:
+            queryset = queryset.filter(user_id=user_id)
+
+        rows = list(
+            queryset.values(
+                "id",
+                "user_id",
+                "session_id",
+                "ts",
+                "event_type",
+                "direction",
+                "origin_transport",
+                "trace_id",
+            )[:limit]
+        )
+        if rows:
+            return rows, "django"
+        try:
+            manticore_rows = get_recent_event_rows(
+                minutes=minutes,
+                service=service,
+                user_id=user_id,
+                limit=limit,
+            )
+        except Exception:
+            manticore_rows = []
+        return manticore_rows, "manticore" if manticore_rows else "django"
+
+    def add_arguments(self, parser):
+        parser.add_argument("--minutes", type=int, default=120)
+        parser.add_argument("--service", default="")
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--limit", type=int, default=200)
+        parser.add_argument("--require-types", default="")
+        parser.add_argument("--fail-if-empty", action="store_true", default=False)
+        parser.add_argument("--json", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        minutes = max(1, int(options.get("minutes") or 120))
+        service = str(options.get("service") or "").strip().lower()
+        user_id = str(options.get("user_id") or "").strip()
+        limit = max(1, int(options.get("limit") or 200))
+        require_types_raw = str(options.get("require_types") or "").strip()
+        fail_if_empty = bool(options.get("fail_if_empty"))
+        as_json = bool(options.get("json"))
+        required_types = [
+            item.strip().lower()
+            for item in require_types_raw.split(",")
+            if item.strip()
+        ]
+
+        rows, data_source = self._recent_rows(
+            minutes=minutes,
+            service=service,
+            user_id=user_id,
+            limit=limit,
+        )
+        event_type_counts = {}
+        for row in rows:
+            key = str(row.get("event_type") or "")
+            event_type_counts[key] = int(event_type_counts.get(key) or 0) + 1
+        missing_required_types = [
+            event_type
+            for event_type in required_types
+            if int(event_type_counts.get(event_type) or 0) <= 0
+        ]
+
+        payload = {
+            "minutes": minutes,
+            "service": service,
+            "user_id": user_id,
+            "data_source": data_source,
+            "count": len(rows),
+            "event_type_counts": event_type_counts,
+            "required_types": required_types,
+            "missing_required_types": missing_required_types,
+            "sample": rows[:25],
+        }
+
+        if as_json:
+            self.stdout.write(json.dumps(payload, indent=2, sort_keys=True))
+            return
+
+        self.stdout.write(
+            f"event-ledger-smoke minutes={minutes} service={service or '-'} user={user_id or '-'} source={data_source} count={len(rows)}"
+        )
+        self.stdout.write(f"event_type_counts={event_type_counts}")
+        if required_types:
+            self.stdout.write(
+                f"required_types={required_types} missing_required_types={missing_required_types}"
+            )
+
+        if fail_if_empty and len(rows) == 0:
+            raise CommandError("No recent canonical event rows found.")
+        if missing_required_types:
+            raise CommandError(
+                "Missing required event types: " + ", ".join(missing_required_types)
+            )
--- a/core/management/commands/event_projection_shadow.py
+++ b/core/management/commands/event_projection_shadow.py
@@ -0,0 +1,136 @@
+from __future__ import annotations
+
+import json
+import time
+
+from django.core.management.base import BaseCommand, CommandError
+
+from core.events.projection import shadow_compare_session
+from core.models import ChatSession, Message
+
+
+class Command(BaseCommand):
+    help = (
+        "Run event->message shadow projection comparison and emit mismatch counters "
+        "per chat session."
+    )
+
+    def add_arguments(self, parser):
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--session-id", default="")
+        parser.add_argument("--service", default="")
+        parser.add_argument("--recent-only", action="store_true", default=False)
+        parser.add_argument("--recent-minutes", type=int, default=0)
+        parser.add_argument("--limit-sessions", type=int, default=50)
+        parser.add_argument("--detail-limit", type=int, default=25)
+        parser.add_argument("--fail-on-mismatch", action="store_true", default=False)
+        parser.add_argument("--json", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        user_id = str(options.get("user_id") or "").strip()
+        session_id = str(options.get("session_id") or "").strip()
+        service = str(options.get("service") or "").strip().lower()
+        recent_only = bool(options.get("recent_only"))
+        recent_minutes = max(0, int(options.get("recent_minutes") or 0))
+        if recent_only and recent_minutes <= 0:
+            recent_minutes = 120
+        limit_sessions = max(1, int(options.get("limit_sessions") or 50))
+        detail_limit = max(0, int(options.get("detail_limit") or 25))
+        as_json = bool(options.get("json"))
+        fail_on_mismatch = bool(options.get("fail_on_mismatch"))
+
+        sessions = ChatSession.objects.all().order_by("-last_interaction", "id")
+        if user_id:
+            sessions = sessions.filter(user_id=user_id)
+        if session_id:
+            sessions = sessions.filter(id=session_id)
+        if service:
+            sessions = sessions.filter(identifier__service=service)
+        if recent_minutes > 0:
+            cutoff_ts = int(time.time() * 1000) - (recent_minutes * 60 * 1000)
+            recent_session_ids = (
+                Message.objects.filter(ts__gte=cutoff_ts)
+                .values_list("session_id", flat=True)
+                .distinct()
+            )
+            sessions = sessions.filter(id__in=recent_session_ids)
+        sessions = list(sessions.select_related("user", "identifier")[:limit_sessions])
+
+        if not sessions:
+            raise CommandError("No chat sessions matched.")
+
+        aggregate = {
+            "sessions_scanned": 0,
+            "db_message_count": 0,
+            "projected_message_count": 0,
+            "mismatch_total": 0,
+            "counters": {
+                "missing_in_projection": 0,
+                "missing_in_db": 0,
+                "text_mismatch": 0,
+                "ts_mismatch": 0,
+                "delivered_ts_mismatch": 0,
+                "read_ts_mismatch": 0,
+                "reactions_mismatch": 0,
+            },
+            "cause_counts": {
+                "missing_event_write": 0,
+                "ambiguous_reaction_target": 0,
+                "payload_normalization_gap": 0,
+            },
+        }
+        results = []
+
+        for session in sessions:
+            compared = shadow_compare_session(session, detail_limit=detail_limit)
+            aggregate["sessions_scanned"] += 1
+            aggregate["db_message_count"] += int(compared.get("db_message_count") or 0)
+            aggregate["projected_message_count"] += int(
+                compared.get("projected_message_count") or 0
+            )
+            aggregate["mismatch_total"] += int(compared.get("mismatch_total") or 0)
+            for key in aggregate["counters"].keys():
+                aggregate["counters"][key] += int(
+                    (compared.get("counters") or {}).get(key) or 0
+                )
+            for key in aggregate["cause_counts"].keys():
+                aggregate["cause_counts"][key] += int(
+                    (compared.get("cause_counts") or {}).get(key) or 0
+                )
+            results.append(compared)
+
+        payload = {
+            "filters": {
+                "user_id": user_id,
+                "session_id": session_id,
+                "service": service,
+                "recent_only": recent_only,
+                "recent_minutes": recent_minutes,
+                "limit_sessions": limit_sessions,
+                "detail_limit": detail_limit,
+            },
+            "aggregate": aggregate,
+            "sessions": results,
+        }
+        if as_json:
+            self.stdout.write(json.dumps(payload, indent=2, sort_keys=True))
+        else:
+            self.stdout.write(
+                "shadow compare: "
+                f"sessions={aggregate['sessions_scanned']} "
+                f"db={aggregate['db_message_count']} "
+                f"projected={aggregate['projected_message_count']} "
+                f"mismatches={aggregate['mismatch_total']}"
+            )
+            self.stdout.write(f"counters={aggregate['counters']}")
+            self.stdout.write(f"cause_counts={aggregate['cause_counts']}")
+            for row in results:
+                self.stdout.write(
+                    f"session={row.get('session_id')} mismatch_total={row.get('mismatch_total')} "
+                    f"db={row.get('db_message_count')} projected={row.get('projected_message_count')}"
+                )
+
+        if fail_on_mismatch and int(aggregate["mismatch_total"] or 0) > 0:
+            raise CommandError(
+                f"Shadow projection mismatch detected: {aggregate['mismatch_total']}"
+            )
--- a/core/management/commands/gia_analysis.py
+++ b/core/management/commands/gia_analysis.py
@@ -0,0 +1,96 @@
+from __future__ import annotations
+
+import time
+
+from django.core.management.base import BaseCommand
+
+from core.events.behavior import summarize_metrics
+from core.events.manticore import get_event_ledger_backend
+from core.util import logs
+
+log = logs.get_logger("gia_analysis")
+
+
+class Command(BaseCommand):
+    help = "Compute behavioral metrics from Manticore event rows into gia_metrics."
+
+    def add_arguments(self, parser):
+        parser.add_argument("--once", action="store_true", default=False)
+        parser.add_argument("--user-id", type=int)
+        parser.add_argument("--person-id")
+        parser.add_argument("--sleep-seconds", type=float, default=60.0)
+        parser.add_argument("--window-days", nargs="*", type=int, default=[1, 7, 30, 90])
+
+    def _run_cycle(
+        self,
+        *,
+        user_id: int | None = None,
+        person_id: str = "",
+        window_days: list[int] | None = None,
+    ) -> int:
+        backend = get_event_ledger_backend()
+        now_ms = int(time.time() * 1000)
+        baseline_since = now_ms - (90 * 86400000)
+        windows = sorted({max(1, int(value)) for value in list(window_days or [1, 7, 30, 90])})
+
+        targets = backend.list_event_targets(user_id=user_id)
+        if person_id:
+            targets = [
+                row
+                for row in targets
+                if str(row.get("person_id") or "").strip() == str(person_id).strip()
+            ]
+
+        written = 0
+        for target in targets:
+            target_user_id = int(target.get("user_id") or 0)
+            target_person_id = str(target.get("person_id") or "").strip()
+            if target_user_id <= 0 or not target_person_id:
+                continue
+            baseline_rows = backend.fetch_events(
+                user_id=target_user_id,
+                person_id=target_person_id,
+                since_ts=baseline_since,
+            )
+            if not baseline_rows:
+                continue
+            for window in windows:
+                since_ts = now_ms - (int(window) * 86400000)
+                window_rows = [
+                    row
+                    for row in baseline_rows
+                    if int(row.get("ts") or 0) >= since_ts
+                ]
+                metrics = summarize_metrics(window_rows, baseline_rows)
+                for metric, values in metrics.items():
+                    backend.upsert_metric(
+                        user_id=target_user_id,
+                        person_id=target_person_id,
+                        window_days=int(window),
+                        metric=metric,
+                        value_ms=int(values.get("value_ms") or 0),
+                        baseline_ms=int(values.get("baseline_ms") or 0),
+                        z_score=float(values.get("z_score") or 0.0),
+                        sample_n=int(values.get("sample_n") or 0),
+                        computed_at=now_ms,
+                    )
+                    written += 1
+        return written
+
+    def handle(self, *args, **options):
+        once = bool(options.get("once"))
+        sleep_seconds = max(1.0, float(options.get("sleep_seconds") or 60.0))
+        user_id = options.get("user_id")
+        person_id = str(options.get("person_id") or "").strip()
+        window_days = list(options.get("window_days") or [1, 7, 30, 90])
+
+        while True:
+            written = self._run_cycle(
+                user_id=user_id,
+                person_id=person_id,
+                window_days=window_days,
+            )
+            self.stdout.write(f"gia-analysis wrote={written}")
+            if once:
+                return
+            time.sleep(sleep_seconds)
--- a/core/management/commands/manticore_backfill.py
+++ b/core/management/commands/manticore_backfill.py
@@ -0,0 +1,46 @@
+from __future__ import annotations
+
+from django.core.management.base import BaseCommand, CommandError
+
+from core.events.manticore import upsert_conversation_event
+from core.models import ConversationEvent
+
+
+class Command(BaseCommand):
+    help = "Backfill behavioral events into Manticore from ConversationEvent rows."
+
+    def add_arguments(self, parser):
+        parser.add_argument(
+            "--from-conversation-events",
+            action="store_true",
+            help="Replay ConversationEvent rows into the Manticore event table.",
+        )
+        parser.add_argument("--user-id", type=int, default=None)
+        parser.add_argument("--limit", type=int, default=5000)
+
+    def handle(self, *args, **options):
+        if not bool(options.get("from_conversation_events")):
+            raise CommandError("Pass --from-conversation-events to run this backfill.")
+
+        queryset = (
+            ConversationEvent.objects.select_related("session__identifier")
+            .order_by("ts", "created_at")
+        )
+        user_id = options.get("user_id")
+        if user_id is not None:
+            queryset = queryset.filter(user_id=int(user_id))
+
+        scanned = 0
+        indexed = 0
+        limit = max(1, int(options.get("limit") or 5000))
+        for event in queryset[:limit]:
+            scanned += 1
+            upsert_conversation_event(event)
+            indexed += 1
+
+        self.stdout.write(
+            self.style.SUCCESS(
+                "manticore-backfill scanned=%s indexed=%s user=%s"
+                % (scanned, indexed, user_id if user_id is not None else "-")
+            )
+        )
--- a/core/management/commands/mcp_manticore_server.py
+++ b/core/management/commands/mcp_manticore_server.py
@@ -0,0 +1,11 @@
+from django.core.management.base import BaseCommand
+
+from core.mcp.server import run_stdio_server
+
+
+class Command(BaseCommand):
+    help = "Run GIA MCP stdio server with manticore/task/documentation tools."
+
+    def handle(self, *args, **options):
+        _ = args, options
+        run_stdio_server()
--- a/core/management/commands/memory_hygiene.py
+++ b/core/management/commands/memory_hygiene.py
@@ -0,0 +1,40 @@
+from __future__ import annotations
+
+import json
+
+from django.core.management.base import BaseCommand
+
+from core.memory.pipeline import run_memory_hygiene
+
+
+class Command(BaseCommand):
+    help = "Run memory hygiene checks (stale decay + contradiction queueing)."
+
+    def add_arguments(self, parser):
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--dry-run", action="store_true", default=False)
+        parser.add_argument("--json", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        user_id_raw = str(options.get("user_id") or "").strip()
+        dry_run = bool(options.get("dry_run"))
+        as_json = bool(options.get("json"))
+        user_id = int(user_id_raw) if user_id_raw else None
+
+        result = run_memory_hygiene(user_id=user_id, dry_run=dry_run)
+        payload = {
+            "user_id": user_id,
+            "dry_run": dry_run,
+            "result": result,
+        }
+        if as_json:
+            self.stdout.write(json.dumps(payload, indent=2, sort_keys=True))
+            return
+        self.stdout.write(
+            "memory-hygiene "
+            f"user={user_id if user_id is not None else '-'} "
+            f"dry_run={'yes' if dry_run else 'no'} "
+            f"expired={int(result.get('expired') or 0)} "
+            f"contradictions={int(result.get('contradictions') or 0)} "
+            f"queued={int(result.get('queued_requests') or 0)}"
+        )
--- a/core/management/commands/memory_search_query.py
+++ b/core/management/commands/memory_search_query.py
@@ -0,0 +1,72 @@
+from __future__ import annotations
+
+import json
+
+from django.core.management.base import BaseCommand, CommandError
+
+from core.memory.search_backend import get_memory_search_backend
+
+
+class Command(BaseCommand):
+    help = "Run a query against configured memory search backend."
+
+    def add_arguments(self, parser):
+        parser.add_argument("--user-id", required=True)
+        parser.add_argument("--query", required=True)
+        parser.add_argument("--conversation-id", default="")
+        parser.add_argument("--statuses", default="active")
+        parser.add_argument("--limit", type=int, default=20)
+        parser.add_argument("--json", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        user_id_raw = str(options.get("user_id") or "").strip()
+        query = str(options.get("query") or "").strip()
+        conversation_id = str(options.get("conversation_id") or "").strip()
+        statuses = tuple(
+            item.strip().lower()
+            for item in str(options.get("statuses") or "active").split(",")
+            if item.strip()
+        )
+        limit = max(1, int(options.get("limit") or 20))
+        as_json = bool(options.get("json"))
+
+        if not user_id_raw:
+            raise CommandError("--user-id is required")
+        if not query:
+            raise CommandError("--query is required")
+
+        backend = get_memory_search_backend()
+        hits = backend.search(
+            user_id=int(user_id_raw),
+            query=query,
+            conversation_id=conversation_id,
+            limit=limit,
+            include_statuses=statuses,
+        )
+        payload = {
+            "backend": getattr(backend, "name", "unknown"),
+            "query": query,
+            "user_id": int(user_id_raw),
+            "conversation_id": conversation_id,
+            "statuses": statuses,
+            "count": len(hits),
+            "hits": [
+                {
+                    "memory_id": item.memory_id,
+                    "score": item.score,
+                    "summary": item.summary,
+                    "payload": item.payload,
+                }
+                for item in hits
+            ],
+        }
+        if as_json:
+            self.stdout.write(json.dumps(payload, indent=2, sort_keys=True))
+            return
+        self.stdout.write(
+            f"memory-search-query backend={payload['backend']} count={payload['count']} query={query!r}"
+        )
+        for row in payload["hits"]:
+            self.stdout.write(
+                f"- id={row['memory_id']} score={row['score']:.2f} summary={row['summary'][:120]}"
+            )
--- a/core/management/commands/memory_search_reindex.py
+++ b/core/management/commands/memory_search_reindex.py
@@ -0,0 +1,49 @@
+from __future__ import annotations
+
+import json
+
+from django.core.management.base import BaseCommand
+
+from core.memory.search_backend import get_memory_search_backend
+
+
+class Command(BaseCommand):
+    help = "Reindex MemoryItem rows into the configured memory search backend."
+
+    def add_arguments(self, parser):
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--statuses", default="active")
+        parser.add_argument("--limit", type=int, default=2000)
+        parser.add_argument("--json", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        user_id_raw = str(options.get("user_id") or "").strip()
+        statuses = tuple(
+            item.strip().lower()
+            for item in str(options.get("statuses") or "active").split(",")
+            if item.strip()
+        )
+        limit = max(1, int(options.get("limit") or 2000))
+        as_json = bool(options.get("json"))
+
+        backend = get_memory_search_backend()
+        result = backend.reindex(
+            user_id=int(user_id_raw) if user_id_raw else None,
+            include_statuses=statuses,
+            limit=limit,
+        )
+        payload = {
+            "backend": getattr(backend, "name", "unknown"),
+            "user_id": user_id_raw,
+            "statuses": statuses,
+            "limit": limit,
+            "result": result,
+        }
+        if as_json:
+            self.stdout.write(json.dumps(payload, indent=2, sort_keys=True))
+            return
+        self.stdout.write(
+            f"memory-search-reindex backend={payload['backend']} "
+            f"user={user_id_raw or '-'} statuses={','.join(statuses) or '-'} "
+            f"scanned={int(result.get('scanned') or 0)} indexed={int(result.get('indexed') or 0)}"
+        )
--- a/core/management/commands/memory_suggest_from_messages.py
+++ b/core/management/commands/memory_suggest_from_messages.py
@@ -0,0 +1,46 @@
+from __future__ import annotations
+
+import json
+
+from django.core.management.base import BaseCommand, CommandError
+
+from core.memory.pipeline import suggest_memories_from_recent_messages
+
+
+class Command(BaseCommand):
+    help = "Suggest proposed MemoryItem rows from recent inbound message text."
+
+    def add_arguments(self, parser):
+        parser.add_argument("--user-id", required=True)
+        parser.add_argument("--limit-messages", type=int, default=300)
+        parser.add_argument("--max-items", type=int, default=30)
+        parser.add_argument("--json", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        user_id_raw = str(options.get("user_id") or "").strip()
+        if not user_id_raw:
+            raise CommandError("--user-id is required")
+        limit_messages = max(1, int(options.get("limit_messages") or 300))
+        max_items = max(1, int(options.get("max_items") or 30))
+        as_json = bool(options.get("json"))
+
+        result = suggest_memories_from_recent_messages(
+            user_id=int(user_id_raw),
+            limit_messages=limit_messages,
+            max_items=max_items,
+        )
+        payload = {
+            "user_id": int(user_id_raw),
+            "limit_messages": limit_messages,
+            "max_items": max_items,
+            "result": result,
+        }
+        if as_json:
+            self.stdout.write(json.dumps(payload, indent=2, sort_keys=True))
+            return
+        self.stdout.write(
+            "memory-suggest-from-messages "
+            f"user={payload['user_id']} "
+            f"scanned={int(result.get('scanned') or 0)} "
+            f"queued={int(result.get('queued') or 0)}"
+        )
--- a/core/management/commands/prune_behavioral_orm_data.py
+++ b/core/management/commands/prune_behavioral_orm_data.py
@@ -0,0 +1,62 @@
+from __future__ import annotations
+
+import time
+
+from django.conf import settings
+from django.core.management.base import BaseCommand
+
+from core.models import ConversationEvent
+
+
+class Command(BaseCommand):
+    help = (
+        "Prune high-growth behavioral ORM shadow tables after data has been "
+        "persisted to Manticore."
+    )
+
+    def add_arguments(self, parser):
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--dry-run", action="store_true", default=False)
+        parser.add_argument("--conversation-days", type=int)
+        parser.add_argument(
+            "--tables",
+            default="conversation_events",
+            help="Comma separated subset of: conversation_events",
+        )
+
+    def _cutoff_ms(self, days: int) -> int:
+        return int(time.time() * 1000) - (max(1, int(days)) * 24 * 60 * 60 * 1000)
+
+    def handle(self, *args, **options):
+        user_id = str(options.get("user_id") or "").strip()
+        dry_run = bool(options.get("dry_run"))
+        conversation_days = int(
+            options.get("conversation_days")
+            or getattr(settings, "CONVERSATION_EVENT_RETENTION_DAYS", 90)
+            or 90
+        )
+        selected_tables = {
+            str(item or "").strip().lower()
+            for item in str(options.get("tables") or "").split(",")
+            if str(item or "").strip()
+        }
+
+        deleted = {
+            "conversation_events": 0,
+        }
+
+        if "conversation_events" in selected_tables:
+            qs = ConversationEvent.objects.filter(
+                ts__lt=self._cutoff_ms(conversation_days)
+            )
+            if user_id:
+                qs = qs.filter(user_id=user_id)
+            deleted["conversation_events"] = int(qs.count() if dry_run else qs.delete()[0])
+
+        self.stdout.write(
+            "prune-behavioral-orm-data "
+            f"dry_run={dry_run} "
+            f"user_id={user_id or '-'} "
+            f"conversation_days={conversation_days} "
+            f"deleted={deleted}"
+        )
--- a/core/management/commands/recalculate_contact_availability.py
+++ b/core/management/commands/recalculate_contact_availability.py
@@ -0,0 +1,161 @@
+from __future__ import annotations
+
+from django.core.management.base import BaseCommand
+
+from core.events.ledger import append_event_sync
+from core.models import Message
+from core.presence.inference import now_ms
+
+
+class Command(BaseCommand):
+    help = (
+        "Replay behavioral event ledger rows from persisted message, receipt, "
+        "and reaction history."
+    )
+
+    def add_arguments(self, parser):
+        parser.add_argument("--days", type=int, default=90)
+        parser.add_argument("--limit", type=int, default=20000)
+        parser.add_argument("--service", default="")
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--dry-run", action="store_true", default=False)
+        parser.add_argument("--no-reset", action="store_true", default=False)
+
+    def _iter_messages(self, *, days: int, limit: int, service: str, user_id: str):
+        cutoff_ts = now_ms() - (max(1, int(days)) * 24 * 60 * 60 * 1000)
+        qs = Message.objects.filter(ts__gte=cutoff_ts).select_related(
+            "user", "session", "session__identifier", "session__identifier__person"
+        )
+        if service:
+            qs = qs.filter(source_service=str(service).strip().lower())
+        if user_id:
+            qs = qs.filter(user_id=str(user_id).strip())
+        return qs.order_by("ts")[: max(1, int(limit))]
+
+    def handle(self, *args, **options):
+        days = max(1, int(options.get("days") or 90))
+        limit = max(1, int(options.get("limit") or 20000))
+        service_filter = str(options.get("service") or "").strip().lower()
+        user_filter = str(options.get("user_id") or "").strip()
+        dry_run = bool(options.get("dry_run"))
+
+        messages = list(
+            self._iter_messages(
+                days=days,
+                limit=limit,
+                service=service_filter,
+                user_id=user_filter,
+            )
+        )
+        indexed = 0
+
+        for msg in messages:
+            session = getattr(msg, "session", None)
+            identifier = getattr(session, "identifier", None)
+            person = getattr(identifier, "person", None)
+            user = getattr(msg, "user", None)
+            if not session or not identifier or not person or not user:
+                continue
+
+            service = (
+                str(getattr(msg, "source_service", "") or identifier.service or "")
+                .strip()
+                .lower()
+            )
+            if not service:
+                continue
+
+            author = str(getattr(msg, "custom_author", "") or "").strip().upper()
+            outgoing = author in {"USER", "BOT"}
+            message_id = str(
+                getattr(msg, "source_message_id", "") or f"django-message-{msg.id}"
+            ).strip()
+
+            if not dry_run:
+                append_event_sync(
+                    user=user,
+                    session=session,
+                    ts=int(getattr(msg, "ts", 0) or 0),
+                    event_type="message_created",
+                    direction="out" if outgoing else "in",
+                    actor_identifier=str(
+                        getattr(msg, "sender_uuid", "") or identifier.identifier or ""
+                    ),
+                    origin_transport=service,
+                    origin_message_id=message_id,
+                    origin_chat_id=str(getattr(msg, "source_chat_id", "") or ""),
+                    payload={
+                        "origin": "recalculate_contact_availability",
+                        "message_id": str(msg.id),
+                        "text": str(getattr(msg, "text", "") or ""),
+                        "outgoing": outgoing,
+                    },
+                )
+            indexed += 1
+
+            read_ts = int(getattr(msg, "read_ts", 0) or 0)
+            if read_ts > 0:
+                if not dry_run:
+                    append_event_sync(
+                        user=user,
+                        session=session,
+                        ts=read_ts,
+                        event_type="read_receipt",
+                        direction="system",
+                        actor_identifier=str(
+                            getattr(msg, "read_by_identifier", "")
+                            or identifier.identifier
+                        ),
+                        origin_transport=service,
+                        origin_message_id=message_id,
+                        origin_chat_id=str(getattr(msg, "source_chat_id", "") or ""),
+                        payload={
+                            "origin": "recalculate_contact_availability",
+                            "message_id": str(msg.id),
+                            "message_ts": int(getattr(msg, "ts", 0) or 0),
+                            "read_by": str(
+                                getattr(msg, "read_by_identifier", "") or ""
+                            ).strip(),
+                        },
+                    )
+                indexed += 1
+
+            reactions = list(
+                (getattr(msg, "receipt_payload", {}) or {}).get("reactions") or []
+            )
+            for reaction in reactions:
+                item = dict(reaction or {})
+                if bool(item.get("removed")):
+                    continue
+                reaction_ts = int(item.get("updated_at") or 0)
+                if reaction_ts <= 0:
+                    continue
+                if not dry_run:
+                    append_event_sync(
+                        user=user,
+                        session=session,
+                        ts=reaction_ts,
+                        event_type="presence_available",
+                        direction="system",
+                        actor_identifier=str(item.get("actor") or ""),
+                        origin_transport=service,
+                        origin_message_id=message_id,
+                        origin_chat_id=str(getattr(msg, "source_chat_id", "") or ""),
+                        payload={
+                            "origin": "recalculate_contact_availability",
+                            "message_id": str(msg.id),
+                            "inferred_from": "reaction",
+                            "emoji": str(item.get("emoji") or ""),
+                            "source_service": str(item.get("source_service") or service),
+                        },
+                    )
+                indexed += 1
+
+        self.stdout.write(
+            self.style.SUCCESS(
+                "recalculate_contact_availability complete "
+                f"messages_scanned={len(messages)} indexed={indexed} "
+                f"dry_run={dry_run} no_reset={bool(options.get('no_reset'))} "
+                f"days={days} limit={limit}"
+            )
+        )
--- a/core/management/commands/reconcile_workspace_metric_history.py
+++ b/core/management/commands/reconcile_workspace_metric_history.py
@@ -0,0 +1,468 @@
+from __future__ import annotations
+
+import statistics
+from datetime import datetime, timezone
+
+from django.core.management.base import BaseCommand
+from django.utils import timezone as dj_timezone
+
+from core.models import (
+    Message,
+    Person,
+    PersonIdentifier,
+    WorkspaceConversation,
+    WorkspaceMetricSnapshot,
+)
+from core.views.workspace import _conversation_for_person
+from core.workspace import compact_snapshot_rows
+
+
+def _score_from_lag(lag_ms, target_hours=4):
+    if lag_ms is None:
+        return 50.0
+    target_ms = max(1, int(target_hours)) * 60 * 60 * 1000
+    return max(0.0, min(100.0, 100.0 / (1.0 + (float(lag_ms) / target_ms))))
+
+
+def _median_or_none(values):
+    if not values:
+        return None
+    return float(statistics.median(values))
+
+
+def _calibrating_payload(last_ts=None):
+    return {
+        "source_event_ts": int(last_ts) if last_ts else None,
+        "stability_state": WorkspaceConversation.StabilityState.CALIBRATING,
+        "stability_score": None,
+        "stability_confidence": 0.0,
+        "stability_sample_messages": 0,
+        "stability_sample_days": 0,
+        "commitment_inbound_score": None,
+        "commitment_outbound_score": None,
+        "commitment_confidence": 0.0,
+        "inbound_messages": 0,
+        "outbound_messages": 0,
+        "reciprocity_score": None,
+        "continuity_score": None,
+        "response_score": None,
+        "volatility_score": None,
+        "inbound_response_score": None,
+        "outbound_response_score": None,
+        "balance_inbound_score": None,
+        "balance_outbound_score": None,
+    }
+
+
+def _compute_payload(rows, identifier_values):
+    if not rows:
+        return _calibrating_payload(None)
+
+    inbound_count = 0
+    outbound_count = 0
+    daily_counts = {}
+    inbound_response_lags = []
+    outbound_response_lags = []
+    pending_in_ts = None
+    pending_out_ts = None
+    first_ts = int(rows[0]["ts"] or 0)
+    last_ts = int(rows[-1]["ts"] or 0)
+    latest_service = (
+        str(rows[-1].get("session__identifier__service") or "").strip().lower()
+    )
+
+    for row in rows:
+        ts = int(row.get("ts") or 0)
+        sender = str(row.get("sender_uuid") or "").strip()
+        author = str(row.get("custom_author") or "").strip().upper()
+        if author in {"USER", "BOT"}:
+            is_inbound = False
+        elif author == "OTHER":
+            is_inbound = True
+        else:
+            is_inbound = sender in identifier_values
+        direction = "in" if is_inbound else "out"
+        day_key = datetime.fromtimestamp(ts / 1000, tz=timezone.utc).date().isoformat()
+        daily_counts[day_key] = daily_counts.get(day_key, 0) + 1
+
+        if direction == "in":
+            inbound_count += 1
+            if pending_out_ts is not None and ts >= pending_out_ts:
+                inbound_response_lags.append(ts - pending_out_ts)
+                pending_out_ts = None
+            pending_in_ts = ts
+        else:
+            outbound_count += 1
+            if pending_in_ts is not None and ts >= pending_in_ts:
+                outbound_response_lags.append(ts - pending_in_ts)
+                pending_in_ts = None
+            pending_out_ts = ts
+
+    message_count = len(rows)
+    span_days = max(1, int(((last_ts - first_ts) / (24 * 60 * 60 * 1000)) + 1))
+    sample_days = len(daily_counts)
+
+    total_messages = max(1, inbound_count + outbound_count)
+    reciprocity_score = 100.0 * (
+        1.0 - abs(inbound_count - outbound_count) / total_messages
+    )
+    continuity_score = 100.0 * min(1.0, sample_days / max(1, span_days))
+    out_resp_score = _score_from_lag(_median_or_none(outbound_response_lags))
+    in_resp_score = _score_from_lag(_median_or_none(inbound_response_lags))
+    response_score = (out_resp_score + in_resp_score) / 2.0
+
+    daily_values = list(daily_counts.values())
+    if len(daily_values) > 1:
+        mean_daily = statistics.mean(daily_values)
+        stdev_daily = statistics.pstdev(daily_values)
+        cv = (stdev_daily / mean_daily) if mean_daily else 1.0
+        volatility_score = max(0.0, 100.0 * (1.0 - min(cv, 1.5) / 1.5))
+    else:
+        volatility_score = 60.0
+
+    stability_score = (
+        (0.35 * reciprocity_score)
+        + (0.25 * continuity_score)
+        + (0.20 * response_score)
+        + (0.20 * volatility_score)
+    )
+
+    balance_out = 100.0 * min(1.0, outbound_count / max(1, inbound_count))
+    balance_in = 100.0 * min(1.0, inbound_count / max(1, outbound_count))
+    commitment_out = (0.60 * out_resp_score) + (0.40 * balance_out)
+    commitment_in = (0.60 * in_resp_score) + (0.40 * balance_in)
+
+    msg_conf = min(1.0, message_count / 200.0)
+    day_conf = min(1.0, sample_days / 30.0)
+    pair_conf = min(
+        1.0, (len(inbound_response_lags) + len(outbound_response_lags)) / 40.0
+    )
+    confidence = (0.50 * msg_conf) + (0.30 * day_conf) + (0.20 * pair_conf)
+
+    if message_count < 20 or sample_days < 3 or confidence < 0.25:
+        stability_state = WorkspaceConversation.StabilityState.CALIBRATING
+        stability_score_value = None
+        commitment_in_value = None
+        commitment_out_value = None
+    else:
+        stability_score_value = round(stability_score, 2)
+        commitment_in_value = round(commitment_in, 2)
+        commitment_out_value = round(commitment_out, 2)
+        if stability_score_value >= 70:
+            stability_state = WorkspaceConversation.StabilityState.STABLE
+        elif stability_score_value >= 50:
+            stability_state = WorkspaceConversation.StabilityState.WATCH
+        else:
+            stability_state = WorkspaceConversation.StabilityState.FRAGILE
+
+    feedback_state = "balanced"
+    if outbound_count > (inbound_count * 1.5):
+        feedback_state = "withdrawing"
+    elif inbound_count > (outbound_count * 1.5):
+        feedback_state = "overextending"
+
+    payload = {
+        "source_event_ts": last_ts,
+        "stability_state": stability_state,
+        "stability_score": (
+            float(stability_score_value) if stability_score_value is not None else None
+        ),
+        "stability_confidence": round(confidence, 3),
+        "stability_sample_messages": message_count,
+        "stability_sample_days": sample_days,
+        "commitment_inbound_score": (
+            float(commitment_in_value) if commitment_in_value is not None else None
+        ),
+        "commitment_outbound_score": (
+            float(commitment_out_value) if commitment_out_value is not None else None
+        ),
+        "commitment_confidence": round(confidence, 3),
+        "inbound_messages": inbound_count,
+        "outbound_messages": outbound_count,
+        "reciprocity_score": round(reciprocity_score, 3),
+        "continuity_score": round(continuity_score, 3),
+        "response_score": round(response_score, 3),
+        "volatility_score": round(volatility_score, 3),
+        "inbound_response_score": round(in_resp_score, 3),
+        "outbound_response_score": round(out_resp_score, 3),
+        "balance_inbound_score": round(balance_in, 3),
+        "balance_outbound_score": round(balance_out, 3),
+    }
+    return payload, latest_service, feedback_state
+
+
+def _payload_signature(payload: dict) -> tuple:
+    return (
+        int(payload.get("source_event_ts") or 0),
+        str(payload.get("stability_state") or ""),
+        payload.get("stability_score"),
+        float(payload.get("stability_confidence") or 0.0),
+        int(payload.get("stability_sample_messages") or 0),
+        int(payload.get("stability_sample_days") or 0),
+        payload.get("commitment_inbound_score"),
+        payload.get("commitment_outbound_score"),
+        float(payload.get("commitment_confidence") or 0.0),
+        int(payload.get("inbound_messages") or 0),
+        int(payload.get("outbound_messages") or 0),
+    )
+
+
+class Command(BaseCommand):
+    help = (
+        "Reconcile AI Workspace metric history by deterministically rebuilding "
+        "WorkspaceMetricSnapshot points from message history."
+    )
+
+    def add_arguments(self, parser):
+        parser.add_argument("--days", type=int, default=365)
+        parser.add_argument("--service", default="")
+        parser.add_argument("--user-id", default="")
+        parser.add_argument("--person-id", default="")
+        parser.add_argument("--step-messages", type=int, default=2)
+        parser.add_argument("--limit", type=int, default=200000)
+        parser.add_argument("--dry-run", action="store_true", default=False)
+        parser.add_argument("--no-reset", action="store_true", default=False)
+        parser.add_argument("--skip-compact", action="store_true", default=False)
+
+    def handle(self, *args, **options):
+        days = max(1, int(options.get("days") or 365))
+        service = str(options.get("service") or "").strip().lower()
+        user_id = str(options.get("user_id") or "").strip()
+        person_id = str(options.get("person_id") or "").strip()
+        step_messages = max(1, int(options.get("step_messages") or 2))
+        limit = max(1, int(options.get("limit") or 200000))
+        dry_run = bool(options.get("dry_run"))
+        reset = not bool(options.get("no_reset"))
+        compact_enabled = not bool(options.get("skip_compact"))
+        today_start = (
+            dj_timezone.now()
+            .astimezone(timezone.utc)
+            .replace(
+                hour=0,
+                minute=0,
+                second=0,
+                microsecond=0,
+            )
+        )
+        cutoff_ts = int((today_start.timestamp() * 1000) - (days * 24 * 60 * 60 * 1000))
+
+        people_qs = Person.objects.all()
+        if user_id:
+            people_qs = people_qs.filter(user_id=user_id)
+        if person_id:
+            people_qs = people_qs.filter(id=person_id)
+        people = list(people_qs.order_by("user_id", "name", "id"))
+
+        conversations_scanned = 0
+        deleted = 0
+        snapshots_created = 0
+        checkpoints_total = 0
+        compacted_deleted = 0
+
+        for person in people:
+            identifiers_qs = PersonIdentifier.objects.filter(
+                user=person.user, person=person
+            )
+            if service:
+                identifiers_qs = identifiers_qs.filter(service=service)
+            identifiers = list(identifiers_qs)
+            if not identifiers:
+                continue
+            identifier_values = {
+                str(row.identifier or "").strip()
+                for row in identifiers
+                if row.identifier
+            }
+            if not identifier_values:
+                continue
+
+            rows = list(
+                Message.objects.filter(
+                    user=person.user,
+                    session__identifier__in=identifiers,
+                    ts__gte=cutoff_ts,
+                )
+                .order_by("ts", "id")
+                .values(
+                    "id",
+                    "ts",
+                    "sender_uuid",
+                    "custom_author",
+                    "session__identifier__service",
+                )[:limit]
+            )
+            if not rows:
+                continue
+
+            conversation = _conversation_for_person(person.user, person)
+            conversations_scanned += 1
+
+            if reset and not dry_run:
+                deleted += WorkspaceMetricSnapshot.objects.filter(
+                    conversation=conversation
+                ).delete()[0]
+
+            existing_signatures = set()
+            if not reset:
+                existing_signatures = set(
+                    _payload_signature(
+                        {
+                            "source_event_ts": row.source_event_ts,
+                            "stability_state": row.stability_state,
+                            "stability_score": row.stability_score,
+                            "stability_confidence": row.stability_confidence,
+                            "stability_sample_messages": row.stability_sample_messages,
+                            "stability_sample_days": row.stability_sample_days,
+                            "commitment_inbound_score": row.commitment_inbound_score,
+                            "commitment_outbound_score": row.commitment_outbound_score,
+                            "commitment_confidence": row.commitment_confidence,
+                            "inbound_messages": row.inbound_messages,
+                            "outbound_messages": row.outbound_messages,
+                        }
+                    )
+                    for row in WorkspaceMetricSnapshot.objects.filter(
+                        conversation=conversation
+                    ).only(
+                        "source_event_ts",
+                        "stability_state",
+                        "stability_score",
+                        "stability_confidence",
+                        "stability_sample_messages",
+                        "stability_sample_days",
+                        "commitment_inbound_score",
+                        "commitment_outbound_score",
+                        "commitment_confidence",
+                        "inbound_messages",
+                        "outbound_messages",
+                    )
+                )
+
+            checkpoints = list(range(step_messages, len(rows) + 1, step_messages))
+            if not checkpoints or checkpoints[-1] != len(rows):
+                checkpoints.append(len(rows))
+            checkpoints_total += len(checkpoints)
+
+            latest_payload = None
+            latest_service = ""
+            latest_feedback_state = "balanced"
+
+            for stop in checkpoints:
+                computed = _compute_payload(rows[:stop], identifier_values)
+                payload = computed[0]
+                latest_payload = payload
+                latest_service = computed[1]
+                latest_feedback_state = computed[2]
+                signature = _payload_signature(payload)
+                if not reset and signature in existing_signatures:
+                    continue
+                snapshots_created += 1
+                if dry_run:
+                    continue
+                WorkspaceMetricSnapshot.objects.create(
+                    conversation=conversation, **payload
+                )
+                existing_signatures.add(signature)
+
+            if not latest_payload:
+                continue
+
+            feedback = dict(conversation.participant_feedback or {})
+            feedback[str(person.id)] = {
+                "state": latest_feedback_state,
+                "inbound_messages": int(latest_payload.get("inbound_messages") or 0),
+                "outbound_messages": int(latest_payload.get("outbound_messages") or 0),
+                "sample_messages": int(
+                    latest_payload.get("stability_sample_messages") or 0
+                ),
+                "sample_days": int(latest_payload.get("stability_sample_days") or 0),
+                "updated_at": dj_timezone.now().isoformat(),
+            }
+            if not dry_run:
+                conversation.platform_type = (
+                    latest_service or conversation.platform_type
+                )
+                conversation.last_event_ts = latest_payload.get("source_event_ts")
+                conversation.stability_state = str(
+                    latest_payload.get("stability_state")
+                    or WorkspaceConversation.StabilityState.CALIBRATING
+                )
+                conversation.stability_score = latest_payload.get("stability_score")
+                conversation.stability_confidence = float(
+                    latest_payload.get("stability_confidence") or 0.0
+                )
+                conversation.stability_sample_messages = int(
+                    latest_payload.get("stability_sample_messages") or 0
+                )
+                conversation.stability_sample_days = int(
+                    latest_payload.get("stability_sample_days") or 0
+                )
+                conversation.commitment_inbound_score = latest_payload.get(
+                    "commitment_inbound_score"
+                )
+                conversation.commitment_outbound_score = latest_payload.get(
+                    "commitment_outbound_score"
+                )
+                conversation.commitment_confidence = float(
+                    latest_payload.get("commitment_confidence") or 0.0
+                )
+                now_ts = dj_timezone.now()
+                conversation.stability_last_computed_at = now_ts
+                conversation.commitment_last_computed_at = now_ts
+                conversation.participant_feedback = feedback
+                conversation.save(
+                    update_fields=[
+                        "platform_type",
+                        "last_event_ts",
+                        "stability_state",
+                        "stability_score",
+                        "stability_confidence",
+                        "stability_sample_messages",
+                        "stability_sample_days",
+                        "stability_last_computed_at",
+                        "commitment_inbound_score",
+                        "commitment_outbound_score",
+                        "commitment_confidence",
+                        "commitment_last_computed_at",
+                        "participant_feedback",
+                    ]
+                )
+                if compact_enabled:
+                    snapshot_rows = list(
+                        WorkspaceMetricSnapshot.objects.filter(
+                            conversation=conversation
+                        )
+                        .order_by("computed_at", "id")
+                        .values("id", "computed_at", "source_event_ts")
+                    )
+                    now_ts_ms = int(dj_timezone.now().timestamp() * 1000)
+                    keep_ids = compact_snapshot_rows(
+                        snapshot_rows=snapshot_rows,
+                        now_ts_ms=now_ts_ms,
+                        cutoff_ts_ms=cutoff_ts,
+                    )
+                    if keep_ids:
+                        compacted_deleted += (
+                            WorkspaceMetricSnapshot.objects.filter(
+                                conversation=conversation
+                            )
+                            .exclude(id__in=list(keep_ids))
+                            .delete()[0]
+                        )
+                    else:
+                        compacted_deleted += WorkspaceMetricSnapshot.objects.filter(
+                            conversation=conversation
+                        ).delete()[0]
+
+        self.stdout.write(
+            self.style.SUCCESS(
+                "reconcile_workspace_metric_history complete "
+                f"conversations_scanned={conversations_scanned} "
+                f"checkpoints={checkpoints_total} "
+                f"created={snapshots_created} "
+                f"deleted={deleted} "
+                f"compacted_deleted={compacted_deleted} "
+                f"compact_enabled={compact_enabled} "
+                f"reset={reset} dry_run={dry_run} "
+                f"days={days} step_messages={step_messages} limit={limit}"
+            )
+        )
--- a/core/management/commands/task_sync_worker.py
+++ b/core/management/commands/task_sync_worker.py
@@ -0,0 +1,9 @@
+from __future__ import annotations
+
+from core.management.commands.codex_worker import Command as LegacyCodexWorkerCommand
+
+
+class Command(LegacyCodexWorkerCommand):
+    help = (
+        "Process queued task-sync events for worker-backed providers (Codex + Claude)."
+    )
--- a/core/mcp/init.py
+++ b/core/mcp/init.py
@@ -0,0 +1,3 @@
+from .server import run_stdio_server
+
+__all__ = ["run_stdio_server"]
--- a/core/mcp/server.py
+++ b/core/mcp/server.py
@@ -0,0 +1,151 @@
+from __future__ import annotations
+
+import json
+import os
+import sys
+from typing import Any
+
+import django
+
+from core.mcp.tools import execute_tool, format_tool_content, tool_specs
+from core.util import logs
+
+log = logs.get_logger("mcp-server")
+
+_compat_newline_mode = False
+
+
+def _setup_django() -> None:
+    os.environ.setdefault("DJANGO_SETTINGS_MODULE", "app.settings")
+    django.setup()
+
+
+def _response(msg_id: Any, result: dict[str, Any]) -> dict[str, Any]:
+    return {"jsonrpc": "2.0", "id": msg_id, "result": result}
+
+
+def _error(msg_id: Any, code: int, message: str) -> dict[str, Any]:
+    return {"jsonrpc": "2.0", "id": msg_id, "error": {"code": code, "message": message}}
+
+
+def _read_message() -> dict[str, Any] | None:
+    global _compat_newline_mode
+    headers: dict[str, str] = {}
+    pending_body = b""
+    while True:
+        line = sys.stdin.buffer.readline()
+        if not line:
+            return None
+        if not headers and line.lstrip().startswith((b"{", b"[")):
+            _compat_newline_mode = True
+            return json.loads(line.decode("utf-8").strip())
+
+        sep = None
+        if b"\r\n\r\n" in line:
+            sep = b"\r\n\r\n"
+        elif b"\n\n" in line:
+            sep = b"\n\n"
+        if sep is not None:
+            header_line, tail = line.split(sep, 1)
+            pending_body = tail
+        else:
+            header_line = line
+
+        if header_line in (b"\r\n", b"\n"):
+            break
+
+        decoded = header_line.decode("utf-8").strip()
+        if ":" in decoded:
+            key, value = decoded.split(":", 1)
+            headers[key.strip().lower()] = value.strip()
+        if sep is not None:
+            break
+
+    length_raw = headers.get("content-length")
+    if not length_raw:
+        if not pending_body:
+            pending_body = sys.stdin.buffer.readline()
+        if not pending_body:
+            return None
+        _compat_newline_mode = True
+        return json.loads(pending_body.decode("utf-8").strip())
+
+    length = int(length_raw)
+    body = pending_body
+    if len(body) < length:
+        body += sys.stdin.buffer.read(length - len(body))
+    body = body[:length]
+    if not body:
+        return None
+    return json.loads(body.decode("utf-8"))
+
+
+def _write_message(payload: dict[str, Any]) -> None:
+    raw_json = json.dumps(payload, separators=(",", ":"), ensure_ascii=False)
+    if _compat_newline_mode:
+        sys.stdout.buffer.write((raw_json + "\n").encode("utf-8"))
+    else:
+        raw = raw_json.encode("utf-8")
+        sys.stdout.buffer.write(f"Content-Length: {len(raw)}\r\n\r\n".encode("utf-8"))
+        sys.stdout.buffer.write(raw)
+    sys.stdout.buffer.flush()
+
+
+def _handle_message(message: dict[str, Any]) -> dict[str, Any] | None:
+    msg_id = message.get("id")
+    method = str(message.get("method") or "")
+    params = message.get("params") or {}
+
+    if method == "notifications/initialized":
+        return None
+    if method == "initialize":
+        return _response(
+            msg_id,
+            {
+                "protocolVersion": "2025-06-18",
+                "serverInfo": {"name": "gia-manticore-mcp", "version": "0.1.0"},
+                "capabilities": {"tools": {}},
+            },
+        )
+    if method == "ping":
+        return _response(msg_id, {})
+    if method == "tools/list":
+        return _response(msg_id, {"tools": tool_specs()})
+    if method == "tools/call":
+        name = str(params.get("name") or "").strip()
+        arguments = params.get("arguments") or {}
+        try:
+            payload = execute_tool(name, arguments)
+            return _response(msg_id, format_tool_content(payload))
+        except Exception as exc:
+            log.warning("tool call failed name=%s err=%s", name, exc)
+            return _response(
+                msg_id,
+                {
+                    "isError": True,
+                    "content": [
+                        {"type": "text", "text": json.dumps({"error": str(exc)})}
+                    ],
+                },
+            )
+
+    return _error(msg_id, -32601, f"Method not found: {method}")
+
+
+def run_stdio_server() -> None:
+    _setup_django()
+    while True:
+        message = _read_message()
+        if message is None:
+            return
+        try:
+            response = _handle_message(message)
+            if response is not None:
+                _write_message(response)
+        except Exception as exc:
+            msg_id = message.get("id")
+            _write_message(_error(msg_id, -32000, str(exc)))
+
+
+if __name__ == "__main__":
+    run_stdio_server()
--- a/core/mcp/tools.py
+++ b/core/mcp/tools.py
--- a/core/memory/init.py
+++ b/core/memory/init.py
@@ -0,0 +1,4 @@
+from .retrieval import retrieve_memories_for_prompt
+from .search_backend import get_memory_search_backend
+
+__all__ = ["get_memory_search_backend", "retrieve_memories_for_prompt"]
--- a/core/memory/pipeline.py
+++ b/core/memory/pipeline.py
@@ -0,0 +1,425 @@
+from __future__ import annotations
+
+import re
+from datetime import timezone as dt_timezone
+from typing import Any
+
+from django.db import transaction
+from django.utils import timezone
+from django.utils.dateparse import parse_datetime
+
+from core.models import (
+    MemoryChangeRequest,
+    MemoryItem,
+    MemorySourceReference,
+    MessageEvent,
+    WorkspaceConversation,
+)
+from core.util import logs
+
+log = logs.get_logger("memory-pipeline")
+
+_LIKE_RE = re.compile(
+    r"\b(?:i (?:like|love|prefer)|my favorite)\s+(?P<value>[^.!?]{2,120})",
+    re.IGNORECASE,
+)
+_DISLIKE_RE = re.compile(
+    r"\b(?:i (?:dislike|hate|avoid)|i don't like)\s+(?P<value>[^.!?]{2,120})",
+    re.IGNORECASE,
+)
+_STYLE_RE = re.compile(
+    r"\b(?:please|pls)\s+(?P<value>[^.!?]{3,120})",
+    re.IGNORECASE,
+)
+
+
+def _clean_value(value: str) -> str:
+    return " ".join(str(value or "").strip().split())
+
+
+def extract_memory_candidates(text: str) -> list[dict[str, Any]]:
+    source = str(text or "").strip()
+    if not source:
+        return []
+
+    candidates: list[dict[str, Any]] = []
+    for regex, field, kind, confidence in (
+        (_LIKE_RE, "likes", "fact", 0.68),
+        (_DISLIKE_RE, "dislikes", "fact", 0.68),
+        (_STYLE_RE, "communication_style", "state", 0.52),
+    ):
+        for match in regex.finditer(source):
+            value = _clean_value(match.group("value"))
+            if len(value) < 3:
+                continue
+            candidates.append(
+                {
+                    "memory_kind": kind,
+                    "field": field,
+                    "text": value,
+                    "confidence_score": confidence,
+                }
+            )
+    return candidates
+
+
+def _existing_fingerprints(user_id: int) -> set[tuple[str, str, str, str]]:
+    items = MemoryItem.objects.filter(user_id=int(user_id)).only(
+        "memory_kind",
+        "conversation_id",
+        "person_id",
+        "content",
+    )
+    fingerprints = set()
+    for item in items:
+        content = item.content or {}
+        field = str(content.get("field") or "").strip().lower()
+        text = _clean_value(str(content.get("text") or "")).lower()
+        fingerprints.add(
+            (
+                str(item.memory_kind or "").strip().lower(),
+                str(item.conversation_id or "").strip(),
+                str(item.person_id or "").strip(),
+                f"{field}:{text}",
+            )
+        )
+    return fingerprints
+
+
+def _infer_single_person_id(conversation: WorkspaceConversation) -> str:
+    participant_ids = list(conversation.participants.values_list("id", flat=True)[:2])
+    if len(participant_ids) != 1:
+        return ""
+    return str(participant_ids[0] or "")
+
+
+@transaction.atomic
+def suggest_memories_from_recent_messages(
+    *,
+    user_id: int,
+    limit_messages: int = 300,
+    max_items: int = 30,
+) -> dict[str, int]:
+    safe_limit_messages = max(1, min(2000, int(limit_messages or 300)))
+    safe_max_items = max(1, min(500, int(max_items or 30)))
+    existing = _existing_fingerprints(int(user_id))
+
+    scanned = 0
+    queued = 0
+    rows = (
+        MessageEvent.objects.filter(user_id=int(user_id), direction="in")
+        .select_related("conversation")
+        .order_by("-ts")[:safe_limit_messages]
+    )
+    for event in rows:
+        scanned += 1
+        person_id = _infer_single_person_id(event.conversation)
+        for candidate in extract_memory_candidates(event.text or ""):
+            field = str(candidate.get("field") or "").strip().lower()
+            text = _clean_value(str(candidate.get("text") or ""))
+            if not text:
+                continue
+            fingerprint = (
+                str(candidate.get("memory_kind") or "fact").strip().lower(),
+                str(event.conversation_id or "").strip(),
+                person_id,
+                f"{field}:{text.lower()}",
+            )
+            if fingerprint in existing:
+                continue
+
+            item = MemoryItem.objects.create(
+                user_id=int(user_id),
+                conversation=event.conversation,
+                person_id=person_id or None,
+                memory_kind=str(candidate.get("memory_kind") or "fact"),
+                status="proposed",
+                content={"field": field, "text": text},
+                provenance={
+                    "pipeline": "message_regex",
+                    "message_event_id": str(event.id),
+                },
+                confidence_score=float(candidate.get("confidence_score") or 0.5),
+            )
+            MemorySourceReference.objects.create(
+                memory=item,
+                message_event=event,
+                source_label="message_event",
+            )
+            MemoryChangeRequest.objects.create(
+                user_id=int(user_id),
+                memory=item,
+                conversation=event.conversation,
+                person_id=person_id or None,
+                action="create",
+                status="pending",
+                proposed_memory_kind=item.memory_kind,
+                proposed_content=item.content,
+                proposed_confidence_score=item.confidence_score,
+                reason="Auto-suggested from recent inbound messages.",
+                requested_by_identifier="memory-pipeline",
+            )
+            existing.add(fingerprint)
+            queued += 1
+            if queued >= safe_max_items:
+                return {"scanned": scanned, "queued": queued}
+    return {"scanned": scanned, "queued": queued}
+
+
+def _coerce_expires_at(value: Any):
+    raw = str(value or "").strip()
+    if not raw:
+        return None
+    parsed = parse_datetime(raw)
+    if parsed is None:
+        raise ValueError("expires_at must be an ISO datetime")
+    if parsed.tzinfo is None:
+        return timezone.make_aware(parsed, dt_timezone.utc)
+    return parsed
+
+
+@transaction.atomic
+def create_memory_change_request(
+    *,
+    user_id: int,
+    action: str,
+    conversation_id: str = "",
+    person_id: str = "",
+    memory_id: str = "",
+    memory_kind: str = "",
+    content: dict[str, Any] | None = None,
+    confidence_score: float | None = None,
+    expires_at: str = "",
+    reason: str = "",
+    requested_by_identifier: str = "",
+) -> MemoryChangeRequest:
+    normalized_action = str(action or "").strip().lower()
+    if normalized_action not in {"create", "update", "delete"}:
+        raise ValueError("action must be create/update/delete")
+
+    memory = None
+    if memory_id:
+        memory = MemoryItem.objects.filter(user_id=int(user_id), id=memory_id).first()
+        if memory is None:
+            raise ValueError("memory_id not found")
+
+    conversation = None
+    if conversation_id:
+        conversation = WorkspaceConversation.objects.filter(
+            user_id=int(user_id),
+            id=conversation_id,
+        ).first()
+        if conversation is None:
+            raise ValueError("conversation_id not found")
+
+    if normalized_action == "create" and conversation is None:
+        raise ValueError("conversation_id is required for create")
+    if normalized_action in {"update", "delete"} and memory is None:
+        raise ValueError("memory_id is required for update/delete")
+
+    return MemoryChangeRequest.objects.create(
+        user_id=int(user_id),
+        memory=memory,
+        conversation=conversation or (memory.conversation if memory else None),
+        person_id=person_id or (str(memory.person_id or "") if memory else "") or None,
+        action=normalized_action,
+        status="pending",
+        proposed_memory_kind=str(
+            memory_kind or (memory.memory_kind if memory else "")
+        ).strip(),
+        proposed_content=dict(content or {}),
+        proposed_confidence_score=(
+            float(confidence_score)
+            if confidence_score is not None
+            else (float(memory.confidence_score) if memory else None)
+        ),
+        proposed_expires_at=_coerce_expires_at(expires_at),
+        reason=str(reason or "").strip(),
+        requested_by_identifier=str(requested_by_identifier or "").strip(),
+    )
+
+
+@transaction.atomic
+def review_memory_change_request(
+    *,
+    user_id: int,
+    request_id: str,
+    decision: str,
+    reviewer_identifier: str = "",
+    note: str = "",
+) -> MemoryChangeRequest:
+    req = MemoryChangeRequest.objects.select_related("memory", "conversation").get(
+        id=request_id,
+        user_id=int(user_id),
+    )
+    if req.status != "pending":
+        raise ValueError("request is not pending")
+
+    now = timezone.now()
+    normalized_decision = str(decision or "").strip().lower()
+    if normalized_decision not in {"approve", "reject"}:
+        raise ValueError("decision must be approve/reject")
+
+    req.reviewed_by_identifier = str(reviewer_identifier or "").strip()
+    req.reviewed_at = now
+    if note:
+        req.reason = f"{req.reason}\n\nReview note: {str(note).strip()}".strip()
+
+    if normalized_decision == "reject":
+        req.status = "rejected"
+        req.save(
+            update_fields=[
+                "status",
+                "reviewed_by_identifier",
+                "reviewed_at",
+                "reason",
+                "updated_at",
+            ]
+        )
+        return req
+
+    req.status = "approved"
+    req.save(
+        update_fields=[
+            "status",
+            "reviewed_by_identifier",
+            "reviewed_at",
+            "reason",
+            "updated_at",
+        ]
+    )
+
+    memory = req.memory
+    if req.action == "create":
+        if memory is None:
+            if req.conversation is None:
+                raise ValueError("approved create request missing conversation")
+            memory = MemoryItem.objects.create(
+                user_id=int(user_id),
+                conversation=req.conversation,
+                person_id=req.person_id,
+                memory_kind=req.proposed_memory_kind or "fact",
+                status="active",
+                content=req.proposed_content or {},
+                confidence_score=float(req.proposed_confidence_score or 0.5),
+                expires_at=req.proposed_expires_at,
+                last_verified_at=now,
+                provenance={"approved_request_id": str(req.id)},
+            )
+            req.memory = memory
+        else:
+            memory.status = "active"
+            memory.last_verified_at = now
+            memory.save(update_fields=["status", "last_verified_at", "updated_at"])
+    elif req.action == "update":
+        if memory is None:
+            raise ValueError("approved update request missing memory")
+        if req.proposed_memory_kind:
+            memory.memory_kind = req.proposed_memory_kind
+        if req.proposed_content:
+            memory.content = req.proposed_content
+        if req.proposed_confidence_score is not None:
+            memory.confidence_score = float(req.proposed_confidence_score)
+        memory.expires_at = req.proposed_expires_at
+        memory.last_verified_at = now
+        memory.status = "active"
+        memory.save()
+    else:
+        if memory is None:
+            raise ValueError("approved delete request missing memory")
+        memory.status = "deprecated"
+        memory.last_verified_at = now
+        memory.save(update_fields=["status", "last_verified_at", "updated_at"])
+
+    req.status = "applied"
+    req.save(update_fields=["status", "memory", "updated_at"])
+    return req
+
+
+@transaction.atomic
+def run_memory_hygiene(
+    *, user_id: int | None = None, dry_run: bool = False
+) -> dict[str, int]:
+    now = timezone.now()
+    queryset = MemoryItem.objects.filter(status="active")
+    if user_id is not None:
+        queryset = queryset.filter(user_id=int(user_id))
+
+    expired_ids = list(
+        queryset.filter(expires_at__isnull=False, expires_at__lte=now).values_list(
+            "id",
+            flat=True,
+        )
+    )
+    expired = len(expired_ids)
+    if expired and not dry_run:
+        MemoryItem.objects.filter(id__in=expired_ids).update(status="deprecated")
+
+    contradictions = 0
+    queued = 0
+    grouped: dict[tuple[int, str, str, str, str], dict[str, list[MemoryItem]]] = {}
+    for item in queryset.select_related("conversation", "person"):
+        content = item.content or {}
+        field = str(content.get("field") or content.get("key") or "").strip().lower()
+        text = _clean_value(
+            str(content.get("text") or content.get("value") or "")
+        ).lower()
+        if not field or not text:
+            continue
+        scope = (
+            int(item.user_id),
+            str(item.person_id or ""),
+            str(item.conversation_id or ""),
+            str(item.memory_kind or ""),
+            field,
+        )
+        grouped.setdefault(scope, {})
+        grouped[scope].setdefault(text, [])
+        grouped[scope][text].append(item)
+
+    for values in grouped.values():
+        if len(values.keys()) <= 1:
+            continue
+        flat = [item for subset in values.values() for item in subset]
+        contradictions += len(flat)
+        if dry_run:
+            continue
+        for item in flat:
+            already_pending = MemoryChangeRequest.objects.filter(
+                user_id=item.user_id,
+                memory=item,
+                action="update",
+                status="pending",
+                reason__icontains="contradiction",
+            ).exists()
+            if already_pending:
+                continue
+            MemoryChangeRequest.objects.create(
+                user_id=item.user_id,
+                memory=item,
+                conversation=item.conversation,
+                person=item.person,
+                action="update",
+                status="pending",
+                proposed_memory_kind=item.memory_kind,
+                proposed_content=item.content,
+                proposed_confidence_score=item.confidence_score,
+                proposed_expires_at=item.expires_at,
+                reason="Contradiction detected by hygiene job.",
+                requested_by_identifier="memory-hygiene",
+            )
+            queued += 1
+
+    log.info(
+        "memory hygiene user=%s dry_run=%s expired=%s contradictions=%s queued=%s",
+        user_id if user_id is not None else "-",
+        dry_run,
+        expired,
+        contradictions,
+        queued,
+    )
+    return {
+        "expired": expired,
+        "contradictions": contradictions,
+        "queued_requests": queued,
+    }
--- a/core/memory/retrieval.py
+++ b/core/memory/retrieval.py
@@ -0,0 +1,133 @@
+from __future__ import annotations
+
+from typing import Any
+
+from django.db.models import Q
+from django.utils import timezone
+
+from core.memory.search_backend import get_memory_search_backend
+from core.models import MemoryItem
+
+
+def _coerce_statuses(value: Any, default: tuple[str, ...]) -> tuple[str, ...]:
+    if isinstance(value, (list, tuple, set)):
+        items = [str(item or "").strip().lower() for item in value]
+    else:
+        items = [item.strip().lower() for item in str(value or "").split(",")]
+    cleaned = tuple(item for item in items if item)
+    return cleaned or default
+
+
+def _base_queryset(
+    *,
+    user_id: int,
+    person_id: str = "",
+    conversation_id: str = "",
+    statuses: tuple[str, ...] = ("active",),
+):
+    now = timezone.now()
+    queryset = MemoryItem.objects.filter(user_id=int(user_id))
+    if statuses:
+        queryset = queryset.filter(status__in=list(statuses))
+    queryset = queryset.filter(Q(expires_at__isnull=True) | Q(expires_at__gt=now))
+    if person_id:
+        queryset = queryset.filter(person_id=person_id)
+    if conversation_id:
+        queryset = queryset.filter(conversation_id=conversation_id)
+    return queryset
+
+
+def retrieve_memories_for_prompt(
+    *,
+    user_id: int,
+    query: str = "",
+    person_id: str = "",
+    conversation_id: str = "",
+    statuses: tuple[str, ...] = ("active",),
+    limit: int = 20,
+) -> list[dict[str, Any]]:
+    statuses = _coerce_statuses(statuses, ("active",))
+    safe_limit = max(1, min(200, int(limit or 20)))
+    search_text = str(query or "").strip()
+
+    if search_text:
+        backend = get_memory_search_backend()
+        hits = backend.search(
+            user_id=int(user_id),
+            query=search_text,
+            conversation_id=conversation_id,
+            limit=safe_limit,
+            include_statuses=statuses,
+        )
+        ids = [
+            str(hit.memory_id or "").strip()
+            for hit in hits
+            if str(hit.memory_id or "").strip()
+        ]
+        scoped = _base_queryset(
+            user_id=int(user_id),
+            person_id=person_id,
+            conversation_id=conversation_id,
+            statuses=statuses,
+        ).filter(id__in=ids)
+        by_id = {str(item.id): item for item in scoped}
+        rows = []
+        for hit in hits:
+            item = by_id.get(str(hit.memory_id))
+            if not item:
+                continue
+            rows.append(
+                {
+                    "id": str(item.id),
+                    "memory_kind": str(item.memory_kind or ""),
+                    "status": str(item.status or ""),
+                    "person_id": str(item.person_id or ""),
+                    "conversation_id": str(item.conversation_id or ""),
+                    "content": item.content or {},
+                    "provenance": item.provenance or {},
+                    "confidence_score": float(item.confidence_score or 0.0),
+                    "expires_at": (
+                        item.expires_at.isoformat() if item.expires_at else ""
+                    ),
+                    "last_verified_at": (
+                        item.last_verified_at.isoformat()
+                        if item.last_verified_at
+                        else ""
+                    ),
+                    "updated_at": (
+                        item.updated_at.isoformat() if item.updated_at else ""
+                    ),
+                    "search_score": float(hit.score or 0.0),
+                    "search_summary": str(hit.summary or ""),
+                }
+            )
+        return rows
+
+    queryset = _base_queryset(
+        user_id=int(user_id),
+        person_id=person_id,
+        conversation_id=conversation_id,
+        statuses=statuses,
+    ).order_by("-last_verified_at", "-updated_at")
+    rows = []
+    for item in queryset[:safe_limit]:
+        rows.append(
+            {
+                "id": str(item.id),
+                "memory_kind": str(item.memory_kind or ""),
+                "status": str(item.status or ""),
+                "person_id": str(item.person_id or ""),
+                "conversation_id": str(item.conversation_id or ""),
+                "content": item.content or {},
+                "provenance": item.provenance or {},
+                "confidence_score": float(item.confidence_score or 0.0),
+                "expires_at": item.expires_at.isoformat() if item.expires_at else "",
+                "last_verified_at": (
+                    item.last_verified_at.isoformat() if item.last_verified_at else ""
+                ),
+                "updated_at": item.updated_at.isoformat() if item.updated_at else "",
+                "search_score": 0.0,
+                "search_summary": "",
+            }
+        )
+    return rows
--- a/core/memory/search_backend.py
+++ b/core/memory/search_backend.py
@@ -0,0 +1,353 @@
+from __future__ import annotations
+
+import hashlib
+import time
+from dataclasses import dataclass
+from typing import Any
+
+import requests
+from django.conf import settings
+
+from core.models import MemoryItem
+from core.util import logs
+
+log = logs.get_logger("memory-search")
+
+
+@dataclass
+class MemorySearchHit:
+    memory_id: str
+    score: float
+    summary: str
+    payload: dict[str, Any]
+
+
+def _flatten_to_text(value: Any) -> str:
+    if value is None:
+        return ""
+    if isinstance(value, dict):
+        parts = []
+        for key, item in value.items():
+            parts.append(str(key))
+            parts.append(_flatten_to_text(item))
+        return " ".join(part for part in parts if part).strip()
+    if isinstance(value, (list, tuple, set)):
+        return " ".join(_flatten_to_text(item) for item in value if item).strip()
+    return str(value).strip()
+
+
+class BaseMemorySearchBackend:
+    def upsert(self, item: MemoryItem) -> None:
+        raise NotImplementedError
+
+    def delete(self, memory_id: str) -> None:
+        raise NotImplementedError
+
+    def search(
+        self,
+        *,
+        user_id: int,
+        query: str,
+        conversation_id: str = "",
+        limit: int = 20,
+        include_statuses: tuple[str, ...] = ("active",),
+    ) -> list[MemorySearchHit]:
+        raise NotImplementedError
+
+    def reindex(
+        self,
+        *,
+        user_id: int | None = None,
+        include_statuses: tuple[str, ...] = ("active",),
+        limit: int = 2000,
+    ) -> dict[str, int]:
+        queryset = MemoryItem.objects.all().order_by("-updated_at")
+        if user_id is not None:
+            queryset = queryset.filter(user_id=int(user_id))
+        if include_statuses:
+            queryset = queryset.filter(status__in=list(include_statuses))
+
+        scanned = 0
+        indexed = 0
+        for item in queryset[: max(1, int(limit))]:
+            scanned += 1
+            try:
+                self.upsert(item)
+                indexed += 1
+            except Exception as exc:
+                log.warning("memory-search upsert failed id=%s err=%s", item.id, exc)
+        return {"scanned": scanned, "indexed": indexed}
+
+
+class DjangoMemorySearchBackend(BaseMemorySearchBackend):
+    name = "django"
+
+    def upsert(self, item: MemoryItem) -> None:
+        # No-op because Django backend queries source-of-truth rows directly.
+        _ = item
+
+    def delete(self, memory_id: str) -> None:
+        _ = memory_id
+
+    def search(
+        self,
+        *,
+        user_id: int,
+        query: str,
+        conversation_id: str = "",
+        limit: int = 20,
+        include_statuses: tuple[str, ...] = ("active",),
+    ) -> list[MemorySearchHit]:
+        needle = str(query or "").strip().lower()
+        if not needle:
+            return []
+
+        queryset = MemoryItem.objects.filter(user_id=int(user_id))
+        if conversation_id:
+            queryset = queryset.filter(conversation_id=conversation_id)
+        if include_statuses:
+            queryset = queryset.filter(status__in=list(include_statuses))
+
+        hits: list[MemorySearchHit] = []
+        for item in queryset.order_by("-updated_at")[:500]:
+            text_blob = _flatten_to_text(item.content).lower()
+            if needle not in text_blob:
+                continue
+            raw_summary = _flatten_to_text(item.content)
+            summary = raw_summary[:280]
+            score = 1.0 + min(1.0, len(needle) / max(1.0, len(text_blob)))
+            hits.append(
+                MemorySearchHit(
+                    memory_id=str(item.id),
+                    score=float(score),
+                    summary=summary,
+                    payload={
+                        "memory_kind": str(item.memory_kind or ""),
+                        "status": str(item.status or ""),
+                        "conversation_id": str(item.conversation_id or ""),
+                        "updated_at": item.updated_at.isoformat(),
+                    },
+                )
+            )
+            if len(hits) >= max(1, int(limit)):
+                break
+        return hits
+
+
+class ManticoreMemorySearchBackend(BaseMemorySearchBackend):
+    name = "manticore"
+    _table_ready_cache: dict[str, float] = {}
+    _table_ready_ttl_seconds = 30.0
+
+    def __init__(self):
+        self.base_url = str(
+            getattr(settings, "MANTICORE_HTTP_URL", "http://localhost:9308")
+        ).rstrip("/")
+        self.table = (
+            str(getattr(settings, "MANTICORE_MEMORY_TABLE", "gia_memory_items")).strip()
+            or "gia_memory_items"
+        )
+        self.timeout_seconds = int(getattr(settings, "MANTICORE_HTTP_TIMEOUT", 5) or 5)
+        self._table_cache_key = f"{self.base_url}|{self.table}"
+
+    def _sql(self, query: str) -> dict[str, Any]:
+        response = requests.post(
+            f"{self.base_url}/sql",
+            data={"mode": "raw", "query": query},
+            timeout=self.timeout_seconds,
+        )
+        response.raise_for_status()
+        payload = response.json()
+        if isinstance(payload, list):
+            return payload[0] if payload else {}
+        return dict(payload or {})
+
+    def ensure_table(self) -> None:
+        last_ready = float(
+            self._table_ready_cache.get(self._table_cache_key, 0.0) or 0.0
+        )
+        if (time.time() - last_ready) <= float(self._table_ready_ttl_seconds):
+            return
+        self._sql(
+            (
+                f"CREATE TABLE IF NOT EXISTS {self.table} ("
+                "id BIGINT,"
+                "memory_uuid STRING,"
+                "user_id BIGINT,"
+                "conversation_id STRING,"
+                "memory_kind STRING,"
+                "status STRING,"
+                "updated_ts BIGINT,"
+                "summary TEXT,"
+                "body TEXT"
+                ")"
+            )
+        )
+        self._table_ready_cache[self._table_cache_key] = time.time()
+
+    def _doc_id(self, memory_id: str) -> int:
+        digest = hashlib.blake2b(
+            str(memory_id or "").encode("utf-8"),
+            digest_size=8,
+        ).digest()
+        value = int.from_bytes(digest, byteorder="big", signed=False)
+        return max(1, int(value))
+
+    def _escape(self, value: Any) -> str:
+        text = str(value or "")
+        text = text.replace("\\", "\\\\").replace("'", "\\'")
+        return text
+
+    def upsert(self, item: MemoryItem) -> None:
+        self.ensure_table()
+        memory_id = str(item.id)
+        doc_id = self._doc_id(memory_id)
+        summary = _flatten_to_text(item.content)[:280]
+        body = _flatten_to_text(item.content)
+        updated_ts = int(item.updated_at.timestamp() * 1000)
+        query = (
+            f"REPLACE INTO {self.table} "
+            "(id,memory_uuid,user_id,conversation_id,memory_kind,status,updated_ts,summary,body) "
+            f"VALUES ({doc_id},'{self._escape(memory_id)}',{int(item.user_id)},"
+            f"'{self._escape(item.conversation_id)}','{self._escape(item.memory_kind)}',"
+            f"'{self._escape(item.status)}',{updated_ts},"
+            f"'{self._escape(summary)}','{self._escape(body)}')"
+        )
+        self._sql(query)
+
+    def _build_upsert_values_clause(self, item: MemoryItem) -> str:
+        memory_id = str(item.id)
+        doc_id = self._doc_id(memory_id)
+        summary = _flatten_to_text(item.content)[:280]
+        body = _flatten_to_text(item.content)
+        updated_ts = int(item.updated_at.timestamp() * 1000)
+        return (
+            f"({doc_id},'{self._escape(memory_id)}',{int(item.user_id)},"
+            f"'{self._escape(item.conversation_id)}','{self._escape(item.memory_kind)}',"
+            f"'{self._escape(item.status)}',{updated_ts},"
+            f"'{self._escape(summary)}','{self._escape(body)}')"
+        )
+
+    def delete(self, memory_id: str) -> None:
+        self.ensure_table()
+        doc_id = self._doc_id(memory_id)
+        self._sql(f"DELETE FROM {self.table} WHERE id={doc_id}")
+
+    def reindex(
+        self,
+        *,
+        user_id: int | None = None,
+        include_statuses: tuple[str, ...] = ("active",),
+        limit: int = 2000,
+    ) -> dict[str, int]:
+        self.ensure_table()
+        queryset = MemoryItem.objects.all().order_by("-updated_at")
+        if user_id is not None:
+            queryset = queryset.filter(user_id=int(user_id))
+        if include_statuses:
+            queryset = queryset.filter(status__in=list(include_statuses))
+
+        scanned = 0
+        indexed = 0
+        batch_size = 100
+        values: list[str] = []
+        for item in queryset[: max(1, int(limit))]:
+            scanned += 1
+            try:
+                values.append(self._build_upsert_values_clause(item))
+            except Exception as exc:
+                log.warning(
+                    "memory-search upsert build failed id=%s err=%s", item.id, exc
+                )
+                continue
+            if len(values) >= batch_size:
+                self._sql(
+                    f"REPLACE INTO {self.table} "
+                    "(id,memory_uuid,user_id,conversation_id,memory_kind,status,updated_ts,summary,body) "
+                    f"VALUES {','.join(values)}"
+                )
+                indexed += len(values)
+                values = []
+        if values:
+            self._sql(
+                f"REPLACE INTO {self.table} "
+                "(id,memory_uuid,user_id,conversation_id,memory_kind,status,updated_ts,summary,body) "
+                f"VALUES {','.join(values)}"
+            )
+            indexed += len(values)
+        return {"scanned": scanned, "indexed": indexed}
+
+    def search(
+        self,
+        *,
+        user_id: int,
+        query: str,
+        conversation_id: str = "",
+        limit: int = 20,
+        include_statuses: tuple[str, ...] = ("active",),
+    ) -> list[MemorySearchHit]:
+        self.ensure_table()
+        needle = str(query or "").strip()
+        if not needle:
+            return []
+
+        where_parts = [f"user_id={int(user_id)}", f"MATCH('{self._escape(needle)}')"]
+        if conversation_id:
+            where_parts.append(f"conversation_id='{self._escape(conversation_id)}'")
+        statuses = [
+            str(item or "").strip()
+            for item in include_statuses
+            if str(item or "").strip()
+        ]
+        if statuses:
+            in_clause = ",".join(f"'{self._escape(item)}'" for item in statuses)
+            where_parts.append(f"status IN ({in_clause})")
+        where_sql = " AND ".join(where_parts)
+        query_sql = (
+            f"SELECT memory_uuid,memory_kind,status,conversation_id,updated_ts,summary,WEIGHT() AS score "
+            f"FROM {self.table} WHERE {where_sql} ORDER BY score DESC LIMIT {max(1, int(limit))}"
+        )
+        payload = self._sql(query_sql)
+        rows = list(payload.get("data") or [])
+        hits = []
+        for row in rows:
+            item = dict(row or {})
+            hits.append(
+                MemorySearchHit(
+                    memory_id=str(item.get("memory_uuid") or ""),
+                    score=float(item.get("score") or 0.0),
+                    summary=str(item.get("summary") or ""),
+                    payload={
+                        "memory_kind": str(item.get("memory_kind") or ""),
+                        "status": str(item.get("status") or ""),
+                        "conversation_id": str(item.get("conversation_id") or ""),
+                        "updated_ts": int(item.get("updated_ts") or 0),
+                    },
+                )
+            )
+        return hits
+
+
+def get_memory_search_backend() -> BaseMemorySearchBackend:
+    backend = str(getattr(settings, "MEMORY_SEARCH_BACKEND", "django")).strip().lower()
+    if backend == "manticore":
+        return ManticoreMemorySearchBackend()
+    return DjangoMemorySearchBackend()
+
+
+def backend_status() -> dict[str, Any]:
+    backend = get_memory_search_backend()
+    status = {
+        "backend": getattr(backend, "name", "unknown"),
+        "ok": True,
+        "ts": int(time.time() * 1000),
+    }
+    if isinstance(backend, ManticoreMemorySearchBackend):
+        try:
+            backend.ensure_table()
+            status["manticore_http_url"] = backend.base_url
+            status["manticore_table"] = backend.table
+        except Exception as exc:
+            status["ok"] = False
+            status["error"] = str(exc)
+    return status
--- a/core/messaging/ai.py
+++ b/core/messaging/ai.py
@@ -1,20 +1,68 @@
+import time
+
+from asgiref.sync import sync_to_async
+from django.utils import timezone
 from openai import AsyncOpenAI

-from core.models import AI
+from core.models import AI, AIRunLog
+
+
+def _prompt_metrics(prompt: list[dict]) -> tuple[int, int]:
+    rows = list(prompt or [])
+    message_count = len(rows)
+    prompt_chars = 0
+    for row in rows:
+        content = ""
+        if isinstance(row, dict):
+            content = str(row.get("content") or "")
+        else:
+            content = str(row or "")
+        prompt_chars += len(content)
+    return message_count, prompt_chars


 async def run_prompt(
-    prompt: list[str],
+    prompt: list[dict],
    ai: AI,
+    operation: str = "",
 ):
+    started_monotonic = time.perf_counter()
+    message_count, prompt_chars = _prompt_metrics(prompt)
+    run_log = await sync_to_async(AIRunLog.objects.create)(
+        user=ai.user,
+        ai=ai,
+        operation=str(operation or "").strip(),
+        model=str(ai.model or ""),
+        base_url=str(ai.base_url or ""),
+        status="running",
+        message_count=message_count,
+        prompt_chars=prompt_chars,
+    )
    cast = {"api_key": ai.api_key}
    if ai.base_url is not None:
        cast["base_url"] = ai.base_url
    client = AsyncOpenAI(**cast)
-    response = await client.chat.completions.create(
-        model=ai.model,
-        messages=prompt,
-    )
-    content = response.choices[0].message.content
-
-    return content
+    try:
+        response = await client.chat.completions.create(
+            model=ai.model,
+            messages=prompt,
+        )
+        content = response.choices[0].message.content
+        duration_ms = int((time.perf_counter() - started_monotonic) * 1000)
+        await sync_to_async(AIRunLog.objects.filter(id=run_log.id).update)(
+            status="ok",
+            response_chars=len(str(content or "")),
+            finished_at=timezone.now(),
+            duration_ms=duration_ms,
+            error="",
+        )
+        return content
+    except Exception as exc:
+        duration_ms = int((time.perf_counter() - started_monotonic) * 1000)
+        await sync_to_async(AIRunLog.objects.filter(id=run_log.id).update)(
+            status="failed",
+            finished_at=timezone.now(),
+            duration_ms=duration_ms,
+            error=str(exc),
+        )
+        raise
--- a/core/messaging/analysis.py
+++ b/core/messaging/analysis.py
@@ -1,6 +1,6 @@
 from django.utils import timezone
-from openai import AsyncOpenAI

+from core.messaging import ai as ai_runner
 from core.models import AI, Manipulation, Person


@@ -50,17 +50,11 @@ def generate_prompt(msg: dict, person: Person, manip: Manipulation, chat_history


 async def run_context_prompt(
-    prompt: list[str],
+    prompt: list[dict],
    ai: AI,
 ):
-    cast = {"api_key": ai.api_key}
-    if ai.base_url is not None:
-        cast["base_url"] = ai.base_url
-    client = AsyncOpenAI(**cast)
-    response = await client.chat.completions.create(
-        model=ai.model,
-        messages=prompt,
+    return await ai_runner.run_prompt(
+        prompt,
+        ai,
+        operation="analysis_context",
    )
-    content = response.choices[0].message.content
-
-    return content
--- a/core/messaging/history.py
+++ b/core/messaging/history.py
@@ -1,8 +1,13 @@
+import time
+import uuid
+
 from asgiref.sync import sync_to_async
 from django.conf import settings

+from core.events.ledger import append_event
 from core.messaging.utils import messages_to_string
 from core.models import ChatSession, Message, QueuedMessage
+from core.observability.tracing import ensure_trace_id
 from core.util import logs

 log = logs.get_logger("history")
@@ -144,7 +149,22 @@ async def get_chat_session(user, identifier):
    return chat_session


-async def store_message(session, sender, text, ts, outgoing=False):
+async def store_message(
+    session,
+    sender,
+    text,
+    ts,
+    outgoing=False,
+    source_service="",
+    source_message_id="",
+    source_chat_id="",
+    reply_to=None,
+    reply_source_service="",
+    reply_source_message_id="",
+    message_meta=None,
+    trace_id="",
+    raw_payload=None,
+):
    log.debug("Storing message for session=%s outgoing=%s", session.id, outgoing)
    msg = await sync_to_async(Message.objects.create)(
        user=session.user,
@@ -154,12 +174,57 @@ async def store_message(session, sender, text, ts, outgoing=False):
        ts=ts,
        delivered_ts=ts,
        custom_author="USER" if outgoing else None,
+        source_service=(source_service or None),
+        source_message_id=str(source_message_id or "").strip() or None,
+        source_chat_id=str(source_chat_id or "").strip() or None,
+        reply_to=reply_to,
+        reply_source_service=str(reply_source_service or "").strip() or None,
+        reply_source_message_id=str(reply_source_message_id or "").strip() or None,
+        message_meta=dict(message_meta or {}),
    )
+    try:
+        await append_event(
+            user=session.user,
+            session=session,
+            ts=int(ts),
+            event_type="message_created",
+            direction="out" if bool(outgoing) else "in",
+            actor_identifier=str(sender or ""),
+            origin_transport=str(source_service or ""),
+            origin_message_id=str(source_message_id or ""),
+            origin_chat_id=str(source_chat_id or ""),
+            payload={
+                "message_id": str(msg.id),
+                "text": str(text or ""),
+                "reply_source_service": str(reply_source_service or ""),
+                "reply_source_message_id": str(reply_source_message_id or ""),
+                "outgoing": bool(outgoing),
+            },
+            raw_payload=dict(raw_payload or {}),
+            trace_id=ensure_trace_id(trace_id, message_meta or {}),
+        )
+    except Exception as exc:
+        log.warning("Event ledger append failed for message=%s: %s", msg.id, exc)

    return msg


-async def store_own_message(session, text, ts, manip=None, queue=False):
+async def store_own_message(
+    session,
+    text,
+    ts,
+    manip=None,
+    queue=False,
+    source_service="",
+    source_message_id="",
+    source_chat_id="",
+    reply_to=None,
+    reply_source_service="",
+    reply_source_message_id="",
+    message_meta=None,
+    trace_id="",
+    raw_payload=None,
+):
    log.debug("Storing own message for session=%s queue=%s", session.id, queue)
    cast = {
        "user": session.user,
@@ -168,6 +233,13 @@ async def store_own_message(session, text, ts, manip=None, queue=False):
        "text": text,
        "ts": ts,
        "delivered_ts": ts,
+        "source_service": (source_service or None),
+        "source_message_id": str(source_message_id or "").strip() or None,
+        "source_chat_id": str(source_chat_id or "").strip() or None,
+        "reply_to": reply_to,
+        "reply_source_service": str(reply_source_service or "").strip() or None,
+        "reply_source_message_id": str(reply_source_message_id or "").strip() or None,
+        "message_meta": dict(message_meta or {}),
    }
    if queue:
        msg_object = QueuedMessage
@@ -178,6 +250,32 @@ async def store_own_message(session, text, ts, manip=None, queue=False):
    msg = await sync_to_async(msg_object.objects.create)(
        **cast,
    )
+    if msg_object is Message:
+        try:
+            await append_event(
+                user=session.user,
+                session=session,
+                ts=int(ts),
+                event_type="message_created",
+                direction="out",
+                actor_identifier="BOT",
+                origin_transport=str(source_service or ""),
+                origin_message_id=str(source_message_id or ""),
+                origin_chat_id=str(source_chat_id or ""),
+                payload={
+                    "message_id": str(msg.id),
+                    "text": str(text or ""),
+                    "queued": bool(queue),
+                    "reply_source_service": str(reply_source_service or ""),
+                    "reply_source_message_id": str(reply_source_message_id or ""),
+                },
+                raw_payload=dict(raw_payload or {}),
+                trace_id=ensure_trace_id(trace_id, message_meta or {}),
+            )
+        except Exception as exc:
+            log.warning(
+                "Event ledger append failed for own message=%s: %s", msg.id, exc
+            )

    return msg

@@ -194,6 +292,8 @@ async def apply_read_receipts(
    source_service="signal",
    read_by_identifier="",
    payload=None,
+    trace_id="",
+    receipt_event_type="read_receipt",
 ):
    """
    Persist delivery/read metadata for one identifier's messages.
@@ -211,6 +311,9 @@ async def apply_read_receipts(
        read_at = int(read_ts) if read_ts else None
    except Exception:
        read_at = None
+    normalized_event_type = str(receipt_event_type or "read_receipt").strip().lower()
+    if normalized_event_type not in {"read_receipt", "delivery_receipt"}:
+        normalized_event_type = "read_receipt"

    rows = await sync_to_async(list)(
        Message.objects.filter(
@@ -225,13 +328,25 @@ async def apply_read_receipts(
        if message.delivered_ts is None:
            message.delivered_ts = read_at or message.ts
            dirty.append("delivered_ts")
-        if read_at and (message.read_ts is None or read_at > message.read_ts):
+        if (
+            normalized_event_type == "read_receipt"
+            and read_at
+            and (message.read_ts is None or read_at > message.read_ts)
+        ):
            message.read_ts = read_at
            dirty.append("read_ts")
-        if source_service and message.read_source_service != source_service:
+        if (
+            normalized_event_type == "read_receipt"
+            and source_service
+            and message.read_source_service != source_service
+        ):
            message.read_source_service = source_service
            dirty.append("read_source_service")
-        if read_by_identifier and message.read_by_identifier != read_by_identifier:
+        if (
+            normalized_event_type == "read_receipt"
+            and read_by_identifier
+            and message.read_by_identifier != read_by_identifier
+        ):
            message.read_by_identifier = read_by_identifier
            dirty.append("read_by_identifier")
        if payload:
@@ -242,6 +357,34 @@ async def apply_read_receipts(
        if dirty:
            await sync_to_async(message.save)(update_fields=dirty)
            updated += 1
+            try:
+                await append_event(
+                    user=user,
+                    session=message.session,
+                    ts=int(read_at or message.ts or 0),
+                    event_type=normalized_event_type,
+                    direction="system",
+                    actor_identifier=str(read_by_identifier or ""),
+                    origin_transport=str(source_service or ""),
+                    origin_message_id=str(message.source_message_id or message.id),
+                    origin_chat_id=str(message.source_chat_id or ""),
+                    payload={
+                        "message_id": str(message.id),
+                        "message_ts": int(message.ts or 0),
+                        "read_ts": int(read_at or 0),
+                        "receipt_event_type": normalized_event_type,
+                        "read_by_identifier": str(read_by_identifier or ""),
+                        "timestamps": [int(v) for v in ts_values],
+                    },
+                    raw_payload=dict(payload or {}),
+                    trace_id=ensure_trace_id(trace_id, payload or {}),
+                )
+            except Exception as exc:
+                log.warning(
+                    "Event ledger append failed for receipt message=%s: %s",
+                    message.id,
+                    exc,
+                )
    return updated


@@ -256,6 +399,8 @@ async def apply_reaction(
    actor="",
    remove=False,
    payload=None,
+    trace_id="",
+    target_author="",
 ):
    log.debug(
        "reaction-bridge history-apply start user=%s person_identifier=%s target_message_id=%s target_ts=%s source=%s actor=%s remove=%s emoji=%s",
@@ -274,11 +419,29 @@ async def apply_reaction(
    ).select_related("session")

    target = None
+    match_strategy = "none"
+    target_author_value = str(target_author or "").strip()
    target_uuid = str(target_message_id or "").strip()
    if target_uuid:
-        target = await sync_to_async(
-            lambda: queryset.filter(id=target_uuid).order_by("-ts").first()
-        )()
+        is_uuid = True
+        try:
+            uuid.UUID(str(target_uuid))
+        except Exception:
+            is_uuid = False
+        if is_uuid:
+            target = await sync_to_async(
+                lambda: queryset.filter(id=target_uuid).order_by("-ts").first()
+            )()
+            if target is not None:
+                match_strategy = "local_message_id"
+        if target is None:
+            target = await sync_to_async(
+                lambda: queryset.filter(source_message_id=target_uuid)
+                .order_by("-ts")
+                .first()
+            )()
+            if target is not None:
+                match_strategy = "source_message_id"

    if target is None:
        try:
@@ -286,11 +449,64 @@ async def apply_reaction(
        except Exception:
            ts_value = 0
        if ts_value > 0:
+            # Signal reactions target source timestamp; prefer deterministic exact matches.
+            exact_candidates = await sync_to_async(list)(
+                queryset.filter(source_message_id=str(ts_value)).order_by("-ts")[:20]
+            )
+            if target_author_value and exact_candidates:
+                filtered = [
+                    row
+                    for row in exact_candidates
+                    if str(row.sender_uuid or "").strip() == target_author_value
+                ]
+                if filtered:
+                    exact_candidates = filtered
+            if exact_candidates:
+                target = exact_candidates[0]
+                match_strategy = "exact_source_message_id_ts"
+                log.debug(
+                    "reaction-bridge history-apply exact-source-ts target_ts=%s picked_message_id=%s candidates=%s",
+                    ts_value,
+                    str(target.id),
+                    len(exact_candidates),
+                )
+
+        if target is None and ts_value > 0:
+            strict_ts_rows = await sync_to_async(list)(
+                queryset.filter(ts=ts_value).order_by("-id")[:20]
+            )
+            if target_author_value and strict_ts_rows:
+                filtered = [
+                    row
+                    for row in strict_ts_rows
+                    if str(row.sender_uuid or "").strip() == target_author_value
+                ]
+                if filtered:
+                    strict_ts_rows = filtered
+            if strict_ts_rows:
+                target = strict_ts_rows[0]
+                match_strategy = "strict_ts_match"
+                log.debug(
+                    "reaction-bridge history-apply strict-ts target_ts=%s picked_message_id=%s candidates=%s",
+                    ts_value,
+                    str(target.id),
+                    len(strict_ts_rows),
+                )
+
+        if target is None and ts_value > 0:
            lower = ts_value - 10_000
            upper = ts_value + 10_000
            window_rows = await sync_to_async(list)(
                queryset.filter(ts__gte=lower, ts__lte=upper).order_by("ts")[:200]
            )
+            if target_author_value and window_rows:
+                author_rows = [
+                    row
+                    for row in window_rows
+                    if str(row.sender_uuid or "").strip() == target_author_value
+                ]
+                if author_rows:
+                    window_rows = author_rows
            if window_rows:
                target = min(
                    window_rows,
@@ -306,6 +522,7 @@ async def apply_reaction(
                    int(target.ts or 0),
                    len(window_rows),
                )
+                match_strategy = "nearest_ts_window"

    if target is None:
        log.warning(
@@ -318,10 +535,13 @@ async def apply_reaction(
        return None

    reactions = list((target.receipt_payload or {}).get("reactions") or [])
+    normalized_source = str(source_service or "").strip().lower()
+    normalized_actor = str(actor or "").strip()
+    normalized_emoji = str(emoji or "").strip()
    reaction_key = (
-        str(source_service or "").strip().lower(),
-        str(actor or "").strip(),
-        str(emoji or "").strip(),
+        normalized_source,
+        normalized_actor,
+        normalized_emoji,
    )

    merged = []
@@ -333,31 +553,94 @@ async def apply_reaction(
            str(row.get("actor") or "").strip(),
            str(row.get("emoji") or "").strip(),
        )
+        if not row_key[2] and bool(row.get("removed")):
+            # Keep malformed remove rows out of active reaction set.
+            continue
        if row_key == reaction_key:
            row["removed"] = bool(remove)
            row["updated_at"] = int(target_ts or target.ts or 0)
            row["payload"] = dict(payload or {})
+            row["match_strategy"] = match_strategy
            merged.append(row)
            replaced = True
            continue
        merged.append(row)

-    if not replaced:
+    if not replaced and (normalized_emoji or not bool(remove)):
        merged.append(
            {
-                "emoji": str(emoji or ""),
-                "source_service": str(source_service or ""),
-                "actor": str(actor or ""),
+                "emoji": normalized_emoji,
+                "source_service": normalized_source,
+                "actor": normalized_actor,
                "removed": bool(remove),
                "updated_at": int(target_ts or target.ts or 0),
                "payload": dict(payload or {}),
+                "match_strategy": match_strategy,
            }
        )
+    elif not replaced and bool(remove):
+        receipt_payload = dict(target.receipt_payload or {})
+        reaction_events = list(receipt_payload.get("reaction_events") or [])
+        reaction_events.append(
+            {
+                "emoji": normalized_emoji,
+                "source_service": normalized_source,
+                "actor": normalized_actor,
+                "removed": True,
+                "updated_at": int(target_ts or target.ts or 0),
+                "payload": dict(payload or {}),
+                "match_strategy": match_strategy,
+                "skip_reason": "remove_without_emoji_or_match",
+            }
+        )
+        if len(reaction_events) > 200:
+            reaction_events = reaction_events[-200:]
+        receipt_payload["reaction_events"] = reaction_events
+        target.receipt_payload = receipt_payload
+        await sync_to_async(target.save)(update_fields=["receipt_payload"])
+        log.debug(
+            "reaction-bridge history-apply remove-without-match message_id=%s strategy=%s",
+            str(target.id),
+            match_strategy,
+        )
+        return target

    receipt_payload = dict(target.receipt_payload or {})
    receipt_payload["reactions"] = merged
+    if match_strategy:
+        receipt_payload["reaction_last_match_strategy"] = str(match_strategy)
    target.receipt_payload = receipt_payload
    await sync_to_async(target.save)(update_fields=["receipt_payload"])
+    try:
+        await append_event(
+            user=user,
+            session=target.session,
+            ts=int(target_ts or target.ts or 0),
+            event_type="reaction_removed" if bool(remove) else "reaction_added",
+            direction="system",
+            actor_identifier=str(actor or ""),
+            origin_transport=str(source_service or ""),
+            origin_message_id=str(target.source_message_id or target.id),
+            origin_chat_id=str(target.source_chat_id or ""),
+            payload={
+                "message_id": str(target.id),
+                "target_message_id": str(target_message_id or target.id),
+                "target_ts": int(target_ts or target.ts or 0),
+                "emoji": str(emoji or ""),
+                "remove": bool(remove),
+                "source_service": normalized_source,
+                "actor": normalized_actor,
+                "match_strategy": match_strategy,
+            },
+            raw_payload=dict(payload or {}),
+            trace_id=ensure_trace_id(trace_id, payload or {}),
+        )
+    except Exception as exc:
+        log.warning(
+            "Event ledger append failed for reaction on message=%s: %s",
+            target.id,
+            exc,
+        )
    log.debug(
        "reaction-bridge history-apply ok message_id=%s reactions=%s",
        str(target.id),
@@ -366,6 +649,277 @@ async def apply_reaction(
    return target


+async def _resolve_message_target(
+    user,
+    identifier,
+    *,
+    target_message_id="",
+    target_ts=0,
+    target_author="",
+):
+    queryset = Message.objects.filter(
+        user=user,
+        session__identifier=identifier,
+    ).select_related("session")
+
+    target = None
+    match_strategy = "none"
+    target_author_value = str(target_author or "").strip()
+    target_uuid = str(target_message_id or "").strip()
+    if target_uuid:
+        is_uuid = True
+        try:
+            uuid.UUID(str(target_uuid))
+        except Exception:
+            is_uuid = False
+        if is_uuid:
+            target = await sync_to_async(
+                lambda: queryset.filter(id=target_uuid).order_by("-ts").first()
+            )()
+            if target is not None:
+                match_strategy = "local_message_id"
+        if target is None:
+            target = await sync_to_async(
+                lambda: queryset.filter(source_message_id=target_uuid)
+                .order_by("-ts")
+                .first()
+            )()
+            if target is not None:
+                match_strategy = "source_message_id"
+
+    if target is None:
+        try:
+            ts_value = int(target_ts or 0)
+        except Exception:
+            ts_value = 0
+        if ts_value > 0:
+            exact_candidates = await sync_to_async(list)(
+                queryset.filter(source_message_id=str(ts_value)).order_by("-ts")[:20]
+            )
+            if target_author_value and exact_candidates:
+                filtered = [
+                    row
+                    for row in exact_candidates
+                    if str(row.sender_uuid or "").strip() == target_author_value
+                ]
+                if filtered:
+                    exact_candidates = filtered
+            if exact_candidates:
+                target = exact_candidates[0]
+                match_strategy = "exact_source_message_id_ts"
+
+        if target is None and ts_value > 0:
+            strict_ts_rows = await sync_to_async(list)(
+                queryset.filter(ts=ts_value).order_by("-id")[:20]
+            )
+            if target_author_value and strict_ts_rows:
+                filtered = [
+                    row
+                    for row in strict_ts_rows
+                    if str(row.sender_uuid or "").strip() == target_author_value
+                ]
+                if filtered:
+                    strict_ts_rows = filtered
+            if strict_ts_rows:
+                target = strict_ts_rows[0]
+                match_strategy = "strict_ts_match"
+
+        if target is None and ts_value > 0:
+            lower = ts_value - 10_000
+            upper = ts_value + 10_000
+            window_rows = await sync_to_async(list)(
+                queryset.filter(ts__gte=lower, ts__lte=upper).order_by("ts")[:200]
+            )
+            if target_author_value and window_rows:
+                author_rows = [
+                    row
+                    for row in window_rows
+                    if str(row.sender_uuid or "").strip() == target_author_value
+                ]
+                if author_rows:
+                    window_rows = author_rows
+            if window_rows:
+                target = min(
+                    window_rows,
+                    key=lambda row: (
+                        abs(int(row.ts or 0) - ts_value),
+                        -int(row.ts or 0),
+                    ),
+                )
+                match_strategy = "nearest_ts_window"
+
+    return target, match_strategy
+
+
+async def apply_message_edit(
+    user,
+    identifier,
+    *,
+    target_message_id="",
+    target_ts=0,
+    new_text="",
+    source_service="",
+    actor="",
+    payload=None,
+    trace_id="",
+    target_author="",
+):
+    target, match_strategy = await _resolve_message_target(
+        user,
+        identifier,
+        target_message_id=target_message_id,
+        target_ts=target_ts,
+        target_author=target_author,
+    )
+    if target is None:
+        log.warning(
+            "edit-sync history-apply miss user=%s person_identifier=%s target_message_id=%s target_ts=%s",
+            getattr(user, "id", "-"),
+            getattr(identifier, "id", "-"),
+            str(target_message_id or "") or "-",
+            int(target_ts or 0),
+        )
+        return None
+
+    old_text = str(target.text or "")
+    updated_text = str(new_text or "")
+    event_ts = int(target_ts or target.ts or int(time.time() * 1000))
+    receipt_payload = dict(target.receipt_payload or {})
+    edit_history = list(receipt_payload.get("edit_history") or [])
+    edit_history.append(
+        {
+            "edited_ts": int(event_ts),
+            "source_service": str(source_service or "").strip().lower(),
+            "actor": str(actor or "").strip(),
+            "previous_text": old_text,
+            "new_text": updated_text,
+            "match_strategy": str(match_strategy or ""),
+            "payload": dict(payload or {}),
+        }
+    )
+    if len(edit_history) > 200:
+        edit_history = edit_history[-200:]
+    receipt_payload["edit_history"] = edit_history
+    receipt_payload["last_edited_ts"] = int(event_ts)
+    receipt_payload["edit_count"] = len(edit_history)
+    target.receipt_payload = receipt_payload
+
+    update_fields = ["receipt_payload"]
+    if old_text != updated_text:
+        target.text = updated_text
+        update_fields.append("text")
+    await sync_to_async(target.save)(update_fields=update_fields)
+    try:
+        await append_event(
+            user=user,
+            session=target.session,
+            ts=int(event_ts),
+            event_type="message_edited",
+            direction="system",
+            actor_identifier=str(actor or ""),
+            origin_transport=str(source_service or ""),
+            origin_message_id=str(target.source_message_id or target.id),
+            origin_chat_id=str(target.source_chat_id or ""),
+            payload={
+                "message_id": str(target.id),
+                "target_message_id": str(target_message_id or target.id),
+                "target_ts": int(target_ts or target.ts or 0),
+                "old_text": old_text,
+                "new_text": updated_text,
+                "source_service": str(source_service or "").strip().lower(),
+                "actor": str(actor or ""),
+                "match_strategy": str(match_strategy or ""),
+            },
+            raw_payload=dict(payload or {}),
+            trace_id=ensure_trace_id(trace_id, payload or {}),
+        )
+    except Exception as exc:
+        log.warning(
+            "Event ledger append failed for message edit message=%s: %s",
+            target.id,
+            exc,
+        )
+    return target
+
+
+async def apply_message_delete(
+    user,
+    identifier,
+    *,
+    target_message_id="",
+    target_ts=0,
+    source_service="",
+    actor="",
+    payload=None,
+    trace_id="",
+    target_author="",
+):
+    target, match_strategy = await _resolve_message_target(
+        user,
+        identifier,
+        target_message_id=target_message_id,
+        target_ts=target_ts,
+        target_author=target_author,
+    )
+    if target is None:
+        log.warning(
+            "delete-sync history-apply miss user=%s person_identifier=%s target_message_id=%s target_ts=%s",
+            getattr(user, "id", "-"),
+            getattr(identifier, "id", "-"),
+            str(target_message_id or "") or "-",
+            int(target_ts or 0),
+        )
+        return None
+
+    event_ts = int(target_ts or target.ts or int(time.time() * 1000))
+    deleted_row = {
+        "deleted_ts": int(event_ts),
+        "source_service": str(source_service or "").strip().lower(),
+        "actor": str(actor or "").strip(),
+        "match_strategy": str(match_strategy or ""),
+        "payload": dict(payload or {}),
+    }
+    receipt_payload = dict(target.receipt_payload or {})
+    delete_events = list(receipt_payload.get("delete_events") or [])
+    delete_events.append(dict(deleted_row))
+    if len(delete_events) > 200:
+        delete_events = delete_events[-200:]
+    receipt_payload["delete_events"] = delete_events
+    receipt_payload["deleted"] = deleted_row
+    receipt_payload["is_deleted"] = True
+    target.receipt_payload = receipt_payload
+    await sync_to_async(target.save)(update_fields=["receipt_payload"])
+    try:
+        await append_event(
+            user=user,
+            session=target.session,
+            ts=int(event_ts),
+            event_type="message_deleted",
+            direction="system",
+            actor_identifier=str(actor or ""),
+            origin_transport=str(source_service or ""),
+            origin_message_id=str(target.source_message_id or target.id),
+            origin_chat_id=str(target.source_chat_id or ""),
+            payload={
+                "message_id": str(target.id),
+                "target_message_id": str(target_message_id or target.id),
+                "target_ts": int(target_ts or target.ts or 0),
+                "source_service": str(source_service or "").strip().lower(),
+                "actor": str(actor or ""),
+                "match_strategy": str(match_strategy or ""),
+            },
+            raw_payload=dict(payload or {}),
+            trace_id=ensure_trace_id(trace_id, payload or {}),
+        )
+    except Exception as exc:
+        log.warning(
+            "Event ledger append failed for message delete message=%s: %s",
+            target.id,
+            exc,
+        )
+    return target
+
+
 def _iter_bridge_refs(receipt_payload, source_service):
    payload = dict(receipt_payload or {})
    refs = payload.get("bridge_refs") or {}
--- a/core/messaging/reply_sync.py
+++ b/core/messaging/reply_sync.py
@@ -0,0 +1,462 @@
+from __future__ import annotations
+
+import re
+from typing import Any
+
+from asgiref.sync import sync_to_async
+
+from core.messaging import history
+from core.models import Message
+
+
+def _as_dict(value: Any) -> dict[str, Any]:
+    return dict(value) if isinstance(value, dict) else {}
+
+
+def _pluck(data: Any, *path: str):
+    cur = data
+    for key in path:
+        if isinstance(cur, dict):
+            cur = cur.get(key)
+            continue
+        if hasattr(cur, key):
+            cur = getattr(cur, key)
+            continue
+        return None
+    return cur
+
+
+def _clean(value: Any) -> str:
+    return str(value or "").strip()
+
+
+def _find_origin_tag(value: Any, depth: int = 0) -> str:
+    if depth > 4:
+        return ""
+    if isinstance(value, dict):
+        direct = _clean(value.get("origin_tag"))
+        if direct:
+            return direct
+        for key in ("metadata", "meta", "message_meta", "contextInfo", "context_info"):
+            nested = _find_origin_tag(value.get(key), depth + 1)
+            if nested:
+                return nested
+        for nested_value in value.values():
+            nested = _find_origin_tag(nested_value, depth + 1)
+            if nested:
+                return nested
+        return ""
+    if isinstance(value, list):
+        for item in value:
+            nested = _find_origin_tag(item, depth + 1)
+            if nested:
+                return nested
+    return ""
+
+
+def _extract_signal_reply(raw_payload: dict[str, Any]) -> dict[str, str]:
+    envelope = _as_dict((raw_payload or {}).get("envelope"))
+    sync_message = _as_dict(envelope.get("syncMessage"))
+    sent_message = _as_dict(sync_message.get("sentMessage"))
+    data_candidates = [
+        _as_dict(envelope.get("dataMessage")),
+        _as_dict(sent_message.get("message")),
+        _as_dict(sent_message),
+        _as_dict((raw_payload or {}).get("dataMessage")),
+        _as_dict(raw_payload),
+    ]
+    quote_key_candidates = (
+        "id",
+        "targetSentTimestamp",
+        "targetTimestamp",
+        "quotedMessageId",
+        "quoted_message_id",
+        "quotedMessageID",
+        "messageId",
+        "message_id",
+        "timestamp",
+    )
+    quote_author_candidates = (
+        "author",
+        "authorUuid",
+        "authorAci",
+        "authorNumber",
+        "source",
+        "sourceNumber",
+        "sourceUuid",
+    )
+    quote_candidates: list[dict[str, Any]] = []
+    for data_message in data_candidates:
+        if not data_message:
+            continue
+        direct_quote = _as_dict(data_message.get("quote") or data_message.get("Quote"))
+        if direct_quote:
+            quote_candidates.append(direct_quote)
+
+        stack = [data_message]
+        while stack:
+            current = stack.pop()
+            if not isinstance(current, dict):
+                continue
+            for key, value in current.items():
+                if isinstance(value, dict):
+                    key_text = str(key or "").strip().lower()
+                    if "quote" in key_text or "reply" in key_text:
+                        quote_candidates.append(_as_dict(value))
+                    stack.append(value)
+                elif isinstance(value, list):
+                    for item in value:
+                        if isinstance(item, dict):
+                            stack.append(item)
+
+    for quote in quote_candidates:
+        quote_id = ""
+        for key in quote_key_candidates:
+            quote_id = _clean(quote.get(key))
+            if quote_id:
+                break
+        if not quote_id:
+            nested = _as_dict(quote.get("id"))
+            if nested:
+                for key in quote_key_candidates:
+                    quote_id = _clean(nested.get(key))
+                    if quote_id:
+                        break
+        if quote_id:
+            reply_chat_id = ""
+            for key in quote_author_candidates:
+                reply_chat_id = _clean(quote.get(key))
+                if reply_chat_id:
+                    break
+            return {
+                "reply_source_message_id": quote_id,
+                "reply_source_service": "signal",
+                "reply_source_chat_id": reply_chat_id,
+            }
+    return {}
+
+
+def _extract_whatsapp_reply(raw_payload: dict[str, Any]) -> dict[str, str]:
+    # Handles common and nested contextInfo/messageContextInfo shapes for
+    # WhatsApp payloads (extended text, media, ephemeral, view-once wrappers).
+    candidate_paths = (
+        ("contextInfo",),
+        ("ContextInfo",),
+        ("messageContextInfo",),
+        ("MessageContextInfo",),
+        ("extendedTextMessage", "contextInfo"),
+        ("ExtendedTextMessage", "ContextInfo"),
+        ("imageMessage", "contextInfo"),
+        ("ImageMessage", "ContextInfo"),
+        ("videoMessage", "contextInfo"),
+        ("VideoMessage", "ContextInfo"),
+        ("documentMessage", "contextInfo"),
+        ("DocumentMessage", "ContextInfo"),
+        ("ephemeralMessage", "message", "contextInfo"),
+        ("ephemeralMessage", "message", "extendedTextMessage", "contextInfo"),
+        ("viewOnceMessage", "message", "contextInfo"),
+        ("viewOnceMessage", "message", "extendedTextMessage", "contextInfo"),
+        ("viewOnceMessageV2", "message", "contextInfo"),
+        ("viewOnceMessageV2", "message", "extendedTextMessage", "contextInfo"),
+        ("viewOnceMessageV2Extension", "message", "contextInfo"),
+        ("viewOnceMessageV2Extension", "message", "extendedTextMessage", "contextInfo"),
+        # snake_case protobuf dict variants
+        ("context_info",),
+        ("message_context_info",),
+        ("extended_text_message", "context_info"),
+        ("image_message", "context_info"),
+        ("video_message", "context_info"),
+        ("document_message", "context_info"),
+        ("ephemeral_message", "message", "context_info"),
+        ("ephemeral_message", "message", "extended_text_message", "context_info"),
+        ("view_once_message", "message", "context_info"),
+        ("view_once_message", "message", "extended_text_message", "context_info"),
+        ("view_once_message_v2", "message", "context_info"),
+        ("view_once_message_v2", "message", "extended_text_message", "context_info"),
+        ("view_once_message_v2_extension", "message", "context_info"),
+        (
+            "view_once_message_v2_extension",
+            "message",
+            "extended_text_message",
+            "context_info",
+        ),
+    )
+    contexts = []
+    for path in candidate_paths:
+        row = _as_dict(_pluck(raw_payload, *path))
+        if row:
+            contexts.append(row)
+    # Recursive fallback for unknown wrapper shapes.
+    stack = [_as_dict(raw_payload)]
+    while stack:
+        current = stack.pop()
+        if not isinstance(current, dict):
+            continue
+        if isinstance(current.get("contextInfo"), dict):
+            contexts.append(_as_dict(current.get("contextInfo")))
+        if isinstance(current.get("ContextInfo"), dict):
+            contexts.append(_as_dict(current.get("ContextInfo")))
+        if isinstance(current.get("messageContextInfo"), dict):
+            contexts.append(_as_dict(current.get("messageContextInfo")))
+        if isinstance(current.get("MessageContextInfo"), dict):
+            contexts.append(_as_dict(current.get("MessageContextInfo")))
+        if isinstance(current.get("context_info"), dict):
+            contexts.append(_as_dict(current.get("context_info")))
+        if isinstance(current.get("message_context_info"), dict):
+            contexts.append(_as_dict(current.get("message_context_info")))
+        for value in current.values():
+            if isinstance(value, dict):
+                stack.append(value)
+            elif isinstance(value, list):
+                for item in value:
+                    if isinstance(item, dict):
+                        stack.append(item)
+
+    for context in contexts:
+        stanza_id = _clean(
+            context.get("stanzaId")
+            or context.get("stanzaID")
+            or context.get("stanza_id")
+            or context.get("StanzaId")
+            or context.get("StanzaID")
+            or context.get("quotedMessageID")
+            or context.get("quotedMessageId")
+            or context.get("QuotedMessageID")
+            or context.get("QuotedMessageId")
+            or _pluck(context, "quotedMessageKey", "id")
+            or _pluck(context, "quoted_message_key", "id")
+            or _pluck(context, "quotedMessage", "key", "id")
+            or _pluck(context, "quoted_message", "key", "id")
+        )
+        if not stanza_id:
+            continue
+        participant = _clean(
+            context.get("participant")
+            or context.get("remoteJid")
+            or context.get("chat")
+            or context.get("Participant")
+            or context.get("RemoteJid")
+            or context.get("RemoteJID")
+            or context.get("Chat")
+        )
+        return {
+            "reply_source_message_id": stanza_id,
+            "reply_source_service": "whatsapp",
+            "reply_source_chat_id": participant,
+        }
+    return {}
+
+
+def extract_whatsapp_reply_debug(raw_payload: dict[str, Any]) -> dict[str, Any]:
+    payload = _as_dict(raw_payload)
+    candidate_paths = (
+        ("contextInfo",),
+        ("ContextInfo",),
+        ("messageContextInfo",),
+        ("MessageContextInfo",),
+        ("extendedTextMessage", "contextInfo"),
+        ("ExtendedTextMessage", "ContextInfo"),
+        ("imageMessage", "contextInfo"),
+        ("ImageMessage", "ContextInfo"),
+        ("videoMessage", "contextInfo"),
+        ("VideoMessage", "ContextInfo"),
+        ("documentMessage", "contextInfo"),
+        ("DocumentMessage", "ContextInfo"),
+        ("ephemeralMessage", "message", "contextInfo"),
+        ("ephemeralMessage", "message", "extendedTextMessage", "contextInfo"),
+        ("viewOnceMessage", "message", "contextInfo"),
+        ("viewOnceMessage", "message", "extendedTextMessage", "contextInfo"),
+        ("viewOnceMessageV2", "message", "contextInfo"),
+        ("viewOnceMessageV2", "message", "extendedTextMessage", "contextInfo"),
+        ("viewOnceMessageV2Extension", "message", "contextInfo"),
+        ("viewOnceMessageV2Extension", "message", "extendedTextMessage", "contextInfo"),
+        ("context_info",),
+        ("message_context_info",),
+        ("extended_text_message", "context_info"),
+        ("image_message", "context_info"),
+        ("video_message", "context_info"),
+        ("document_message", "context_info"),
+        ("ephemeral_message", "message", "context_info"),
+        ("ephemeral_message", "message", "extended_text_message", "context_info"),
+        ("view_once_message", "message", "context_info"),
+        ("view_once_message", "message", "extended_text_message", "context_info"),
+        ("view_once_message_v2", "message", "context_info"),
+        ("view_once_message_v2", "message", "extended_text_message", "context_info"),
+        ("view_once_message_v2_extension", "message", "context_info"),
+        (
+            "view_once_message_v2_extension",
+            "message",
+            "extended_text_message",
+            "context_info",
+        ),
+    )
+    rows = []
+    for path in candidate_paths:
+        context = _as_dict(_pluck(payload, *path))
+        if not context:
+            continue
+        rows.append(
+            {
+                "path": ".".join(path),
+                "keys": sorted([str(key) for key in context.keys()])[:40],
+                "stanzaId": _clean(
+                    context.get("stanzaId")
+                    or context.get("stanzaID")
+                    or context.get("stanza_id")
+                    or context.get("StanzaId")
+                    or context.get("StanzaID")
+                    or context.get("quotedMessageID")
+                    or context.get("quotedMessageId")
+                    or context.get("QuotedMessageID")
+                    or context.get("QuotedMessageId")
+                    or _pluck(context, "quotedMessageKey", "id")
+                    or _pluck(context, "quoted_message_key", "id")
+                    or _pluck(context, "quotedMessage", "key", "id")
+                    or _pluck(context, "quoted_message", "key", "id")
+                ),
+                "participant": _clean(
+                    context.get("participant")
+                    or context.get("remoteJid")
+                    or context.get("chat")
+                    or context.get("Participant")
+                    or context.get("RemoteJid")
+                    or context.get("RemoteJID")
+                    or context.get("Chat")
+                ),
+            }
+        )
+    return {
+        "candidate_count": len(rows),
+        "candidates": rows[:20],
+    }
+
+
+def extract_reply_ref(service: str, raw_payload: dict[str, Any]) -> dict[str, str]:
+    svc = _clean(service).lower()
+    payload = _as_dict(raw_payload)
+    if svc == "xmpp":
+        reply_id = _clean(
+            payload.get("reply_source_message_id") or payload.get("reply_id")
+        )
+        reply_chat = _clean(
+            payload.get("reply_source_chat_id") or payload.get("reply_chat_id")
+        )
+        if reply_id:
+            return {
+                "reply_source_message_id": reply_id,
+                "reply_source_service": "xmpp",
+                "reply_source_chat_id": reply_chat,
+            }
+        return {}
+    if svc == "signal":
+        return _extract_signal_reply(payload)
+    if svc == "whatsapp":
+        return _extract_whatsapp_reply(payload)
+    if svc == "web":
+        reply_id = _clean(payload.get("reply_to_message_id"))
+        if reply_id:
+            return {
+                "reply_source_message_id": reply_id,
+                "reply_source_service": "web",
+                "reply_source_chat_id": _clean(payload.get("reply_source_chat_id")),
+            }
+    return {}
+
+
+def extract_origin_tag(raw_payload: dict[str, Any] | None) -> str:
+    return _find_origin_tag(_as_dict(raw_payload))
+
+
+async def resolve_reply_target(
+    user, session, reply_ref: dict[str, str]
+) -> Message | None:
+    if not reply_ref or session is None:
+        return None
+    reply_source_message_id = _clean(reply_ref.get("reply_source_message_id"))
+    if not reply_source_message_id:
+        return None
+
+    # Direct local UUID fallback (web compose references local Message IDs).
+    if re.fullmatch(
+        r"[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}",
+        reply_source_message_id,
+    ):
+        direct = await sync_to_async(
+            lambda: Message.objects.filter(
+                user=user,
+                session=session,
+                id=reply_source_message_id,
+            ).first()
+        )()
+        if direct is not None:
+            return direct
+
+    source_service = _clean(reply_ref.get("reply_source_service"))
+    by_source = await sync_to_async(
+        lambda: Message.objects.filter(
+            user=user,
+            session=session,
+            source_service=source_service or None,
+            source_message_id=reply_source_message_id,
+        )
+        .order_by("-ts")
+        .first()
+    )()
+    if by_source is not None:
+        return by_source
+
+    # Bridge ref fallback: resolve replies against bridge mappings persisted in
+    # message receipt payloads.
+    identifier = getattr(session, "identifier", None)
+    if identifier is not None:
+        service_candidates = []
+        if source_service:
+            service_candidates.append(source_service)
+        # XMPP replies can target bridged messages from any external service.
+        if source_service == "xmpp":
+            service_candidates.extend(["signal", "whatsapp", "instagram"])
+        for candidate in service_candidates:
+            bridge = await history.resolve_bridge_ref(
+                user=user,
+                identifier=identifier,
+                source_service=candidate,
+                xmpp_message_id=reply_source_message_id,
+                upstream_message_id=reply_source_message_id,
+            )
+            local_message_id = _clean((bridge or {}).get("local_message_id"))
+            if not local_message_id:
+                continue
+            bridged = await sync_to_async(
+                lambda: Message.objects.filter(
+                    user=user,
+                    session=session,
+                    id=local_message_id,
+                ).first()
+            )()
+            if bridged is not None:
+                return bridged
+
+    fallback = await sync_to_async(
+        lambda: Message.objects.filter(
+            user=user,
+            session=session,
+            reply_source_message_id=reply_source_message_id,
+        )
+        .order_by("-ts")
+        .first()
+    )()
+    return fallback
+
+
+def apply_sync_origin(message_meta: dict | None, origin_tag: str) -> dict:
+    payload = dict(message_meta or {})
+    tag = _clean(origin_tag)
+    if not tag:
+        return payload
+    payload["origin_tag"] = tag
+    return payload
+
+
+def is_mirrored_origin(message_meta: dict | None) -> bool:
+    payload = dict(message_meta or {})
+    return bool(_clean(payload.get("origin_tag")))
--- a/core/messaging/text_export.py
+++ b/core/messaging/text_export.py
@@ -0,0 +1,18 @@
+from __future__ import annotations
+
+from typing import Iterable
+
+from core.models import Message
+
+
+def normalize_message_text(message: Message) -> str:
+    text = str(getattr(message, "text", "") or "").strip()
+    return text or "(no text)"
+
+
+def plain_text_lines(messages: Iterable[Message]) -> list[str]:
+    return [normalize_message_text(message) for message in list(messages)]
+
+
+def plain_text_blob(messages: Iterable[Message]) -> str:
+    return "\n".join(plain_text_lines(messages))
--- a/core/migrations/0026_platformchatlink_is_group.py
+++ b/core/migrations/0026_platformchatlink_is_group.py
@@ -0,0 +1,24 @@
+# Generated by Django 5.2.11 on 2026-02-19 14:47
+
+import django.db.models.deletion
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('core', '0025_platformchatlink'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='platformchatlink',
+            name='is_group',
+            field=models.BooleanField(default=False),
+        ),
+        migrations.AlterField(
+            model_name='platformchatlink',
+            name='person',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.CASCADE, to='core.person'),
+        ),
+    ]
--- a/core/migrations/0027_businessplandocument_businessplanrevision_and_more.py
+++ b/core/migrations/0027_businessplandocument_businessplanrevision_and_more.py
@@ -0,0 +1,311 @@
+# Generated by Django 5.2.7 on 2026-03-01 20:03
+
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('core', '0026_platformchatlink_is_group'),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name='BusinessPlanDocument',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('source_service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('source_channel_identifier', models.CharField(blank=True, default='', max_length=255)),
+                ('title', models.CharField(default='Business Plan', max_length=255)),
+                ('status', models.CharField(choices=[('draft', 'Draft'), ('final', 'Final')], default='draft', max_length=32)),
+                ('content_markdown', models.TextField(blank=True, default='')),
+                ('structured_payload', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='BusinessPlanRevision',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('content_markdown', models.TextField(blank=True, default='')),
+                ('structured_payload', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+            ],
+            options={
+                'ordering': ['created_at'],
+            },
+        ),
+        migrations.CreateModel(
+            name='CommandAction',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('action_type', models.CharField(choices=[('extract_bp', 'Extract Business Plan'), ('post_result', 'Post Result'), ('save_document', 'Save Document')], max_length=64)),
+                ('enabled', models.BooleanField(default=True)),
+                ('config', models.JSONField(blank=True, default=dict)),
+                ('position', models.PositiveIntegerField(default=0)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+            options={
+                'ordering': ['position', 'id'],
+            },
+        ),
+        migrations.CreateModel(
+            name='CommandChannelBinding',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('direction', models.CharField(choices=[('ingress', 'Ingress'), ('egress', 'Egress'), ('scratchpad_mirror', 'Scratchpad Mirror')], max_length=64)),
+                ('service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('channel_identifier', models.CharField(max_length=255)),
+                ('enabled', models.BooleanField(default=True)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='CommandProfile',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('slug', models.CharField(default='bp', max_length=64)),
+                ('name', models.CharField(default='Business Plan', max_length=255)),
+                ('enabled', models.BooleanField(default=True)),
+                ('trigger_token', models.CharField(default='#bp#', max_length=64)),
+                ('reply_required', models.BooleanField(default=True)),
+                ('exact_match_only', models.BooleanField(default=True)),
+                ('window_scope', models.CharField(choices=[('conversation', 'Conversation')], default='conversation', max_length=64)),
+                ('template_text', models.TextField(blank=True, default='')),
+                ('visibility_mode', models.CharField(choices=[('status_in_source', 'Status In Source'), ('silent', 'Silent')], default='status_in_source', max_length=64)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='CommandRun',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('status', models.CharField(choices=[('pending', 'Pending'), ('running', 'Running'), ('ok', 'OK'), ('failed', 'Failed'), ('skipped', 'Skipped')], default='pending', max_length=32)),
+                ('error', models.TextField(blank=True, default='')),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TranslationBridge',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('name', models.CharField(default='Translation Bridge', max_length=255)),
+                ('enabled', models.BooleanField(default=True)),
+                ('a_service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('a_channel_identifier', models.CharField(max_length=255)),
+                ('a_language', models.CharField(default='en', max_length=64)),
+                ('b_service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('b_channel_identifier', models.CharField(max_length=255)),
+                ('b_language', models.CharField(default='en', max_length=64)),
+                ('direction', models.CharField(choices=[('a_to_b', 'A To B'), ('b_to_a', 'B To A'), ('bidirectional', 'Bidirectional')], default='bidirectional', max_length=32)),
+                ('quick_mode_title', models.CharField(blank=True, default='', max_length=255)),
+                ('settings', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TranslationEventLog',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('target_service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('target_channel', models.CharField(max_length=255)),
+                ('status', models.CharField(choices=[('pending', 'Pending'), ('ok', 'OK'), ('failed', 'Failed'), ('skipped', 'Skipped')], default='pending', max_length=32)),
+                ('error', models.TextField(blank=True, default='')),
+                ('origin_tag', models.CharField(blank=True, default='', max_length=255)),
+                ('content_hash', models.CharField(blank=True, default='', max_length=255)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='message_meta',
+            field=models.JSONField(blank=True, default=dict, help_text='Normalized message metadata such as origin tags.'),
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='reply_source_message_id',
+            field=models.CharField(blank=True, help_text='Source message id for the replied target.', max_length=255, null=True),
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='reply_source_service',
+            field=models.CharField(blank=True, choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], help_text='Source service for the replied target.', max_length=255, null=True),
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='reply_to',
+            field=models.ForeignKey(blank=True, help_text='Resolved local message this message replies to.', null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='reply_children', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='source_chat_id',
+            field=models.CharField(blank=True, help_text='Source service chat or thread identifier when available.', max_length=255, null=True),
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='source_message_id',
+            field=models.CharField(blank=True, help_text='Source service message id when available.', max_length=255, null=True),
+        ),
+        migrations.AddField(
+            model_name='message',
+            name='source_service',
+            field=models.CharField(blank=True, choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], help_text='Source service where this message originally appeared.', max_length=255, null=True),
+        ),
+        migrations.AddIndex(
+            model_name='message',
+            index=models.Index(fields=['user', 'source_service', 'source_message_id'], name='core_messag_user_id_252699_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='message',
+            index=models.Index(fields=['user', 'session', 'ts'], name='core_messag_user_id_ba0e73_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='message',
+            index=models.Index(fields=['user', 'reply_source_service', 'reply_source_message_id'], name='core_messag_user_id_70ca93_idx'),
+        ),
+        migrations.AddField(
+            model_name='businessplandocument',
+            name='anchor_message',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='business_plan_anchor_docs', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='businessplandocument',
+            name='trigger_message',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='business_plan_trigger_docs', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='businessplandocument',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='businessplanrevision',
+            name='document',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='revisions', to='core.businessplandocument'),
+        ),
+        migrations.AddField(
+            model_name='businessplanrevision',
+            name='editor_user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='commandprofile',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='commandchannelbinding',
+            name='profile',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='channel_bindings', to='core.commandprofile'),
+        ),
+        migrations.AddField(
+            model_name='commandaction',
+            name='profile',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='actions', to='core.commandprofile'),
+        ),
+        migrations.AddField(
+            model_name='businessplandocument',
+            name='command_profile',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='business_plan_documents', to='core.commandprofile'),
+        ),
+        migrations.AddField(
+            model_name='commandrun',
+            name='profile',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='runs', to='core.commandprofile'),
+        ),
+        migrations.AddField(
+            model_name='commandrun',
+            name='result_ref',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='command_runs', to='core.businessplandocument'),
+        ),
+        migrations.AddField(
+            model_name='commandrun',
+            name='trigger_message',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='command_runs', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='commandrun',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='translationbridge',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='translationeventlog',
+            name='bridge',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='events', to='core.translationbridge'),
+        ),
+        migrations.AddField(
+            model_name='translationeventlog',
+            name='source_message',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='translation_events', to='core.message'),
+        ),
+        migrations.AddConstraint(
+            model_name='commandprofile',
+            constraint=models.UniqueConstraint(fields=('user', 'slug'), name='unique_command_profile_per_user'),
+        ),
+        migrations.AddIndex(
+            model_name='commandchannelbinding',
+            index=models.Index(fields=['profile', 'direction', 'service'], name='core_comman_profile_6c16d5_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='commandchannelbinding',
+            index=models.Index(fields=['profile', 'service', 'channel_identifier'], name='core_comman_profile_2c801d_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='commandaction',
+            index=models.Index(fields=['profile', 'action_type', 'enabled'], name='core_comman_profile_f8e752_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='businessplandocument',
+            index=models.Index(fields=['user', 'status', 'updated_at'], name='core_busine_user_id_028f36_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='businessplandocument',
+            index=models.Index(fields=['user', 'source_service', 'source_channel_identifier'], name='core_busine_user_id_54ef14_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='commandrun',
+            index=models.Index(fields=['user', 'status', 'updated_at'], name='core_comman_user_id_aa2881_idx'),
+        ),
+        migrations.AddConstraint(
+            model_name='commandrun',
+            constraint=models.UniqueConstraint(fields=('profile', 'trigger_message'), name='unique_command_run_profile_trigger_message'),
+        ),
+        migrations.AddIndex(
+            model_name='translationbridge',
+            index=models.Index(fields=['user', 'enabled'], name='core_transl_user_id_ce99cd_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='translationbridge',
+            index=models.Index(fields=['user', 'a_service', 'a_channel_identifier'], name='core_transl_user_id_2f26ee_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='translationbridge',
+            index=models.Index(fields=['user', 'b_service', 'b_channel_identifier'], name='core_transl_user_id_1f910a_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='translationeventlog',
+            index=models.Index(fields=['bridge', 'created_at'], name='core_transl_bridge__509ffc_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='translationeventlog',
+            index=models.Index(fields=['bridge', 'status', 'updated_at'], name='core_transl_bridge__0a7676_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='translationeventlog',
+            index=models.Index(fields=['origin_tag'], name='core_transl_origin__a5c2f3_idx'),
+        ),
+    ]
--- a/core/migrations/0028_airunlog.py
+++ b/core/migrations/0028_airunlog.py
@@ -0,0 +1,86 @@
+# Generated by Django 5.2.7 on 2026-03-02 00:00
+
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("core", "0027_businessplandocument_businessplanrevision_and_more"),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="AIRunLog",
+            fields=[
+                (
+                    "id",
+                    models.BigAutoField(
+                        auto_created=True,
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="ID",
+                    ),
+                ),
+                ("operation", models.CharField(blank=True, default="", max_length=64)),
+                ("model", models.CharField(blank=True, default="", max_length=255)),
+                ("base_url", models.CharField(blank=True, default="", max_length=255)),
+                (
+                    "status",
+                    models.CharField(
+                        choices=[("running", "Running"), ("ok", "OK"), ("failed", "Failed")],
+                        default="running",
+                        max_length=16,
+                    ),
+                ),
+                ("message_count", models.PositiveIntegerField(default=0)),
+                ("prompt_chars", models.PositiveIntegerField(default=0)),
+                ("response_chars", models.PositiveIntegerField(default=0)),
+                ("error", models.TextField(blank=True, default="")),
+                ("started_at", models.DateTimeField(auto_now_add=True)),
+                ("finished_at", models.DateTimeField(blank=True, null=True)),
+                ("duration_ms", models.PositiveIntegerField(blank=True, null=True)),
+                (
+                    "ai",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="run_logs",
+                        to="core.ai",
+                    ),
+                ),
+                (
+                    "user",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="ai_run_logs",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
+            ],
+        ),
+        migrations.AddIndex(
+            model_name="airunlog",
+            index=models.Index(fields=["user", "started_at"], name="core_airunl_user_id_6f4700_idx"),
+        ),
+        migrations.AddIndex(
+            model_name="airunlog",
+            index=models.Index(
+                fields=["user", "status", "started_at"],
+                name="core_airunl_user_id_b4486e_idx",
+            ),
+        ),
+        migrations.AddIndex(
+            model_name="airunlog",
+            index=models.Index(
+                fields=["user", "operation", "started_at"],
+                name="core_airunl_user_id_4f0f5e_idx",
+            ),
+        ),
+        migrations.AddIndex(
+            model_name="airunlog",
+            index=models.Index(fields=["user", "model", "started_at"], name="core_airunl_user_id_953bff_idx"),
+        ),
+    ]
--- a/core/migrations/0029_answermemory_answersuggestionevent_chattasksource_and_more.py
+++ b/core/migrations/0029_answermemory_answersuggestionevent_chattasksource_and_more.py
@@ -0,0 +1,342 @@
+# Generated by Django 5.2.11 on 2026-03-02 11:55
+
+import uuid
+
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('core', '0028_airunlog'),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name='AnswerMemory',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('channel_identifier', models.CharField(max_length=255)),
+                ('question_fingerprint', models.CharField(max_length=128)),
+                ('question_text', models.TextField(blank=True, default='')),
+                ('answer_text', models.TextField(blank=True, default='')),
+                ('confidence_meta', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='AnswerSuggestionEvent',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('status', models.CharField(choices=[('suggested', 'Suggested'), ('accepted', 'Accepted'), ('dismissed', 'Dismissed')], default='suggested', max_length=32)),
+                ('score', models.FloatField(default=0.0)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='ChatTaskSource',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('channel_identifier', models.CharField(max_length=255)),
+                ('enabled', models.BooleanField(default=True)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='DerivedTask',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('title', models.CharField(max_length=255)),
+                ('source_service', models.CharField(choices=[('signal', 'Signal'), ('whatsapp', 'WhatsApp'), ('xmpp', 'XMPP'), ('instagram', 'Instagram'), ('web', 'Web')], max_length=255)),
+                ('source_channel', models.CharField(max_length=255)),
+                ('reference_code', models.CharField(blank=True, default='', max_length=64)),
+                ('external_key', models.CharField(blank=True, default='', max_length=255)),
+                ('status_snapshot', models.CharField(blank=True, default='open', max_length=64)),
+                ('immutable_payload', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='DerivedTaskEvent',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('event_type', models.CharField(choices=[('created', 'Created'), ('progress', 'Progress'), ('completion_marked', 'Completion Marked'), ('synced', 'Synced'), ('sync_failed', 'Sync Failed'), ('parse_warning', 'Parse Warning')], max_length=32)),
+                ('actor_identifier', models.CharField(blank=True, default='', max_length=255)),
+                ('payload', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+            ],
+            options={
+                'ordering': ['created_at', 'id'],
+            },
+        ),
+        migrations.CreateModel(
+            name='ExternalSyncEvent',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('provider', models.CharField(default='mock', max_length=64)),
+                ('idempotency_key', models.CharField(blank=True, default='', max_length=255)),
+                ('status', models.CharField(choices=[('pending', 'Pending'), ('ok', 'OK'), ('failed', 'Failed'), ('retrying', 'Retrying')], default='pending', max_length=32)),
+                ('payload', models.JSONField(blank=True, default=dict)),
+                ('error', models.TextField(blank=True, default='')),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TaskCompletionPattern',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('phrase', models.CharField(max_length=64)),
+                ('enabled', models.BooleanField(default=True)),
+                ('position', models.PositiveIntegerField(default=0)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TaskEpic',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('name', models.CharField(max_length=255)),
+                ('external_key', models.CharField(blank=True, default='', max_length=255)),
+                ('active', models.BooleanField(default=True)),
+                ('settings', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TaskProject',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('name', models.CharField(max_length=255)),
+                ('external_key', models.CharField(blank=True, default='', max_length=255)),
+                ('active', models.BooleanField(default=True)),
+                ('settings', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TaskProviderConfig',
+            fields=[
+                ('id', models.UUIDField(default=uuid.uuid4, editable=False, primary_key=True, serialize=False)),
+                ('provider', models.CharField(default='mock', max_length=64)),
+                ('enabled', models.BooleanField(default=False)),
+                ('settings', models.JSONField(blank=True, default=dict)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('updated_at', models.DateTimeField(auto_now=True)),
+            ],
+        ),
+        migrations.RenameIndex(
+            model_name='airunlog',
+            new_name='core_airunl_user_id_13b24a_idx',
+            old_name='core_airunl_user_id_6f4700_idx',
+        ),
+        migrations.RenameIndex(
+            model_name='airunlog',
+            new_name='core_airunl_user_id_678025_idx',
+            old_name='core_airunl_user_id_b4486e_idx',
+        ),
+        migrations.RenameIndex(
+            model_name='airunlog',
+            new_name='core_airunl_user_id_55c2d4_idx',
+            old_name='core_airunl_user_id_4f0f5e_idx',
+        ),
+        migrations.RenameIndex(
+            model_name='airunlog',
+            new_name='core_airunl_user_id_bef024_idx',
+            old_name='core_airunl_user_id_953bff_idx',
+        ),
+        migrations.AddField(
+            model_name='answermemory',
+            name='answer_message',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='answer_memory_rows', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='answermemory',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='answer_memory', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='answersuggestionevent',
+            name='candidate_answer',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='suggestion_events', to='core.answermemory'),
+        ),
+        migrations.AddField(
+            model_name='answersuggestionevent',
+            name='message',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='answer_suggestion_events', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='answersuggestionevent',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='answer_suggestion_events', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='chattasksource',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='chat_task_sources', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='derivedtask',
+            name='origin_message',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='derived_task_origins', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='derivedtask',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='derived_tasks', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='derivedtaskevent',
+            name='source_message',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='derived_task_events', to='core.message'),
+        ),
+        migrations.AddField(
+            model_name='derivedtaskevent',
+            name='task',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='events', to='core.derivedtask'),
+        ),
+        migrations.AddField(
+            model_name='externalsyncevent',
+            name='task',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='external_sync_events', to='core.derivedtask'),
+        ),
+        migrations.AddField(
+            model_name='externalsyncevent',
+            name='task_event',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='external_sync_events', to='core.derivedtaskevent'),
+        ),
+        migrations.AddField(
+            model_name='externalsyncevent',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='external_sync_events', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='taskcompletionpattern',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='task_completion_patterns', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='derivedtask',
+            name='epic',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='derived_tasks', to='core.taskepic'),
+        ),
+        migrations.AddField(
+            model_name='chattasksource',
+            name='epic',
+            field=models.ForeignKey(blank=True, null=True, on_delete=django.db.models.deletion.SET_NULL, related_name='chat_sources', to='core.taskepic'),
+        ),
+        migrations.AddField(
+            model_name='taskproject',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='task_projects', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddField(
+            model_name='taskepic',
+            name='project',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='epics', to='core.taskproject'),
+        ),
+        migrations.AddField(
+            model_name='derivedtask',
+            name='project',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='derived_tasks', to='core.taskproject'),
+        ),
+        migrations.AddField(
+            model_name='chattasksource',
+            name='project',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='chat_sources', to='core.taskproject'),
+        ),
+        migrations.AddField(
+            model_name='taskproviderconfig',
+            name='user',
+            field=models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='task_provider_configs', to=settings.AUTH_USER_MODEL),
+        ),
+        migrations.AddIndex(
+            model_name='answermemory',
+            index=models.Index(fields=['user', 'service', 'channel_identifier', 'created_at'], name='core_answer_user_id_b88ba6_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='answermemory',
+            index=models.Index(fields=['user', 'question_fingerprint', 'created_at'], name='core_answer_user_id_9353c7_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='answersuggestionevent',
+            index=models.Index(fields=['user', 'status', 'created_at'], name='core_answer_user_id_05d0f9_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='answersuggestionevent',
+            index=models.Index(fields=['message', 'status'], name='core_answer_message_1cb119_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='derivedtaskevent',
+            index=models.Index(fields=['task', 'created_at'], name='core_derive_task_id_897ae5_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='derivedtaskevent',
+            index=models.Index(fields=['event_type', 'created_at'], name='core_derive_event_t_1cf04b_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='externalsyncevent',
+            index=models.Index(fields=['user', 'provider', 'status', 'updated_at'], name='core_extern_user_id_e71276_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='externalsyncevent',
+            index=models.Index(fields=['idempotency_key'], name='core_extern_idempot_dce064_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='taskcompletionpattern',
+            index=models.Index(fields=['user', 'enabled', 'position'], name='core_taskco_user_id_0c1b5e_idx'),
+        ),
+        migrations.AddConstraint(
+            model_name='taskcompletionpattern',
+            constraint=models.UniqueConstraint(fields=('user', 'phrase'), name='unique_task_completion_phrase_per_user'),
+        ),
+        migrations.AddIndex(
+            model_name='taskproject',
+            index=models.Index(fields=['user', 'active', 'updated_at'], name='core_taskpr_user_id_4f8472_idx'),
+        ),
+        migrations.AddConstraint(
+            model_name='taskproject',
+            constraint=models.UniqueConstraint(fields=('user', 'name'), name='unique_task_project_name_per_user'),
+        ),
+        migrations.AddIndex(
+            model_name='taskepic',
+            index=models.Index(fields=['project', 'active', 'updated_at'], name='core_taskep_project_ea76c3_idx'),
+        ),
+        migrations.AddConstraint(
+            model_name='taskepic',
+            constraint=models.UniqueConstraint(fields=('project', 'name'), name='unique_task_epic_name_per_project'),
+        ),
+        migrations.AddIndex(
+            model_name='derivedtask',
+            index=models.Index(fields=['user', 'project', 'created_at'], name='core_derive_user_id_a98675_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='derivedtask',
+            index=models.Index(fields=['user', 'source_service', 'source_channel'], name='core_derive_user_id_aaa167_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='derivedtask',
+            index=models.Index(fields=['user', 'reference_code'], name='core_derive_user_id_d06303_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='chattasksource',
+            index=models.Index(fields=['user', 'service', 'channel_identifier', 'enabled'], name='core_chatta_user_id_01f271_idx'),
+        ),
+        migrations.AddIndex(
+            model_name='chattasksource',
+            index=models.Index(fields=['project', 'enabled'], name='core_chatta_project_826bab_idx'),
+        ),
+        migrations.AddConstraint(
+            model_name='taskproviderconfig',
+            constraint=models.UniqueConstraint(fields=('user', 'provider'), name='unique_task_provider_config_per_user'),
+        ),
+    ]
--- a/core/migrations/0030_chattasksource_settings.py
+++ b/core/migrations/0030_chattasksource_settings.py
@@ -0,0 +1,18 @@
+# Generated by Django 5.2.11 on 2026-03-02 12:55
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('core', '0029_answermemory_answersuggestionevent_chattasksource_and_more'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='chattasksource',
+            name='settings',
+            field=models.JSONField(blank=True, default=dict),
+        ),
+    ]
--- a/core/migrations/0031_commandvariantpolicy.py
+++ b/core/migrations/0031_commandvariantpolicy.py
@@ -0,0 +1,107 @@
+# Generated by Django 5.2.11 on 2026-03-02 14:17
+
+import django.db.models.deletion
+from django.db import migrations, models
+
+
+def _backfill_variant_policies(apps, schema_editor):
+    CommandProfile = apps.get_model("core", "CommandProfile")
+    CommandAction = apps.get_model("core", "CommandAction")
+    CommandVariantPolicy = apps.get_model("core", "CommandVariantPolicy")
+
+    for profile in CommandProfile.objects.all().iterator():
+        actions = list(CommandAction.objects.filter(profile=profile))
+        post_result_enabled = any(
+            str(getattr(row, "action_type", "")) == "post_result"
+            and bool(getattr(row, "enabled", False))
+            for row in actions
+        )
+        send_status_to_source = (
+            str(getattr(profile, "visibility_mode", "") or "") == "status_in_source"
+        )
+        if str(getattr(profile, "slug", "") or "") == "bp":
+            rows = (
+                ("bp", "ai", 0),
+                ("bp_set", "verbatim", 1),
+                ("bp_set_range", "verbatim", 2),
+            )
+        else:
+            rows = (("default", "verbatim", 0),)
+
+        for key, generation_mode, position in rows:
+            CommandVariantPolicy.objects.get_or_create(
+                profile=profile,
+                variant_key=key,
+                defaults={
+                    "enabled": True,
+                    "generation_mode": generation_mode,
+                    "send_plan_to_egress": bool(post_result_enabled),
+                    "send_status_to_source": bool(send_status_to_source),
+                    "send_status_to_egress": False,
+                    "position": int(position),
+                },
+            )
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ("core", "0030_chattasksource_settings"),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="CommandVariantPolicy",
+            fields=[
+                (
+                    "id",
+                    models.BigAutoField(
+                        auto_created=True,
+                        primary_key=True,
+                        serialize=False,
+                        verbose_name="ID",
+                    ),
+                ),
+                ("variant_key", models.CharField(default="default", max_length=64)),
+                ("enabled", models.BooleanField(default=True)),
+                (
+                    "generation_mode",
+                    models.CharField(
+                        choices=[("ai", "AI"), ("verbatim", "Verbatim")],
+                        default="verbatim",
+                        max_length=32,
+                    ),
+                ),
+                ("send_plan_to_egress", models.BooleanField(default=True)),
+                ("send_status_to_source", models.BooleanField(default=True)),
+                ("send_status_to_egress", models.BooleanField(default=False)),
+                ("position", models.PositiveIntegerField(default=0)),
+                ("created_at", models.DateTimeField(auto_now_add=True)),
+                ("updated_at", models.DateTimeField(auto_now=True)),
+                (
+                    "profile",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="variant_policies",
+                        to="core.commandprofile",
+                    ),
+                ),
+            ],
+            options={
+                "ordering": ["position", "id"],
+                "indexes": [
+                    models.Index(
+                        fields=["profile", "enabled", "variant_key"],
+                        name="core_comman_profile_7913f5_idx",
+                    )
+                ],
+                "constraints": [
+                    models.UniqueConstraint(
+                        fields=("profile", "variant_key"),
+                        name="unique_command_variant_policy_per_profile",
+                    )
+                ],
+            },
+        ),
+        migrations.RunPython(_backfill_variant_policies, migrations.RunPython.noop),
+    ]
--- a/core/migrations/0032_commandvariantpolicy_store_document.py
+++ b/core/migrations/0032_commandvariantpolicy_store_document.py
@@ -0,0 +1,18 @@
+# Generated by Django 5.2.11 on 2026-03-02 17:38
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('core', '0031_commandvariantpolicy'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='commandvariantpolicy',
+            name='store_document',
+            field=models.BooleanField(default=True),
+        ),
+    ]
--- a/core/migrations/0033_contactavailability_and_externalchatlink.py
+++ b/core/migrations/0033_contactavailability_and_externalchatlink.py
@@ -0,0 +1,236 @@
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ("core", "0032_commandvariantpolicy_store_document"),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name="ContactAvailabilitySettings",
+            fields=[
+                ("id", models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name="ID")),
+                ("enabled", models.BooleanField(default=True)),
+                ("show_in_chat", models.BooleanField(default=True)),
+                ("show_in_groups", models.BooleanField(default=True)),
+                ("inference_enabled", models.BooleanField(default=True)),
+                ("retention_days", models.PositiveIntegerField(default=90)),
+                ("fade_threshold_seconds", models.PositiveIntegerField(default=900)),
+                ("created_at", models.DateTimeField(auto_now_add=True)),
+                ("updated_at", models.DateTimeField(auto_now=True)),
+                (
+                    "user",
+                    models.OneToOneField(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="contact_availability_settings",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
+            ],
+        ),
+        migrations.CreateModel(
+            name="ContactAvailabilityEvent",
+            fields=[
+                ("id", models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name="ID")),
+                (
+                    "service",
+                    models.CharField(
+                        choices=[("signal", "Signal"), ("whatsapp", "WhatsApp"), ("xmpp", "XMPP"), ("instagram", "Instagram"), ("web", "Web")],
+                        max_length=255,
+                    ),
+                ),
+                (
+                    "source_kind",
+                    models.CharField(
+                        choices=[
+                            ("native_presence", "Native Presence"),
+                            ("read_receipt", "Read Receipt"),
+                            ("typing_start", "Typing Start"),
+                            ("typing_stop", "Typing Stop"),
+                            ("message_in", "Message In"),
+                            ("message_out", "Message Out"),
+                            ("inferred_timeout", "Inferred Timeout"),
+                        ],
+                        max_length=32,
+                    ),
+                ),
+                (
+                    "availability_state",
+                    models.CharField(
+                        choices=[("available", "Available"), ("unavailable", "Unavailable"), ("unknown", "Unknown"), ("fading", "Fading")],
+                        max_length=32,
+                    ),
+                ),
+                ("confidence", models.FloatField(default=0.0)),
+                ("ts", models.BigIntegerField(db_index=True)),
+                ("payload", models.JSONField(blank=True, default=dict)),
+                ("created_at", models.DateTimeField(auto_now_add=True)),
+                (
+                    "person",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="availability_events",
+                        to="core.person",
+                    ),
+                ),
+                (
+                    "person_identifier",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="availability_events",
+                        to="core.personidentifier",
+                    ),
+                ),
+                (
+                    "user",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="contact_availability_events",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
+            ],
+            options={
+                "indexes": [
+                    models.Index(fields=["user", "person", "ts"], name="core_contac_user_id_0da9b2_idx"),
+                    models.Index(fields=["user", "service", "ts"], name="core_contac_user_id_bce271_idx"),
+                    models.Index(fields=["user", "availability_state", "ts"], name="core_contac_user_id_1b50b3_idx"),
+                ],
+            },
+        ),
+        migrations.CreateModel(
+            name="ExternalChatLink",
+            fields=[
+                ("id", models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name="ID")),
+                ("provider", models.CharField(default="codex_cli", max_length=64)),
+                ("external_chat_id", models.CharField(max_length=255)),
+                ("metadata", models.JSONField(blank=True, default=dict)),
+                ("enabled", models.BooleanField(default=True)),
+                ("created_at", models.DateTimeField(auto_now_add=True)),
+                ("updated_at", models.DateTimeField(auto_now=True)),
+                (
+                    "person",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="external_chat_links",
+                        to="core.person",
+                    ),
+                ),
+                (
+                    "person_identifier",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="external_chat_links",
+                        to="core.personidentifier",
+                    ),
+                ),
+                (
+                    "user",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="external_chat_links",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
+            ],
+            options={
+                "indexes": [
+                    models.Index(fields=["user", "provider", "external_chat_id"], name="core_extern_user_id_f4a7b0_idx"),
+                    models.Index(fields=["user", "provider", "enabled"], name="core_extern_user_id_7d2295_idx"),
+                ],
+                "constraints": [
+                    models.UniqueConstraint(fields=("user", "provider", "external_chat_id"), name="unique_external_chat_link_per_provider"),
+                ],
+            },
+        ),
+        migrations.CreateModel(
+            name="ContactAvailabilitySpan",
+            fields=[
+                ("id", models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name="ID")),
+                (
+                    "service",
+                    models.CharField(
+                        choices=[("signal", "Signal"), ("whatsapp", "WhatsApp"), ("xmpp", "XMPP"), ("instagram", "Instagram"), ("web", "Web")],
+                        max_length=255,
+                    ),
+                ),
+                (
+                    "state",
+                    models.CharField(
+                        choices=[("available", "Available"), ("unavailable", "Unavailable"), ("unknown", "Unknown"), ("fading", "Fading")],
+                        max_length=32,
+                    ),
+                ),
+                ("start_ts", models.BigIntegerField(db_index=True)),
+                ("end_ts", models.BigIntegerField(db_index=True)),
+                ("confidence_start", models.FloatField(default=0.0)),
+                ("confidence_end", models.FloatField(default=0.0)),
+                ("payload", models.JSONField(blank=True, default=dict)),
+                ("created_at", models.DateTimeField(auto_now_add=True)),
+                ("updated_at", models.DateTimeField(auto_now=True)),
+                (
+                    "closing_event",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="closing_spans",
+                        to="core.contactavailabilityevent",
+                    ),
+                ),
+                (
+                    "opening_event",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="opening_spans",
+                        to="core.contactavailabilityevent",
+                    ),
+                ),
+                (
+                    "person",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="availability_spans",
+                        to="core.person",
+                    ),
+                ),
+                (
+                    "person_identifier",
+                    models.ForeignKey(
+                        blank=True,
+                        null=True,
+                        on_delete=django.db.models.deletion.SET_NULL,
+                        related_name="availability_spans",
+                        to="core.personidentifier",
+                    ),
+                ),
+                (
+                    "user",
+                    models.ForeignKey(
+                        on_delete=django.db.models.deletion.CASCADE,
+                        related_name="contact_availability_spans",
+                        to=settings.AUTH_USER_MODEL,
+                    ),
+                ),
+            ],
+            options={
+                "indexes": [
+                    models.Index(fields=["user", "person", "start_ts"], name="core_contac_user_id_9cd15a_idx"),
+                    models.Index(fields=["user", "person", "end_ts"], name="core_contac_user_id_88584a_idx"),
+                    models.Index(fields=["user", "service", "start_ts"], name="core_contac_user_id_182ffb_idx"),
+                ],
+            },
+        ),
+    ]
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Mark Veidemanis	79766d279d	Begin reimplementing compose	2026-03-12 00:12:57 +00:00
Mark Veidemanis	1570f79b62	Re-theme	2026-03-11 15:12:17 +00:00
Mark Veidemanis	cbedcd67f6	Implement Manticore fully and re-theme	2026-03-11 02:19:08 +00:00
Mark Veidemanis	da044be68c	Improve cross-platform group integration	2026-03-09 12:28:30 +00:00
Mark Veidemanis	a7421b9350	Fix some task quirks	2026-03-08 23:16:15 +00:00
Mark Veidemanis	acedc01e83	Fix all integrations	2026-03-08 22:08:55 +00:00
Mark Veidemanis	bca4d6898f	Increase security and reformat	2026-03-07 20:52:13 +00:00
Mark Veidemanis	10588a18b9	Improve settings hierarchy conciseness	2026-03-07 16:32:24 +00:00
Mark Veidemanis	611de57bf8	Improve security	2026-03-07 15:34:23 +00:00
Mark Veidemanis	add685a326	Remove real contact numbers from tests and update tooling - Replace real phone numbers in tests with Ofcom-reserved fictitious numbers (447700900xxx range) throughout test suite - Add SIGNAL_NUMBER to stack.env.example documenting required env var - Update pre-commit hooks to latest versions (black 26.3.0, isort 8.0.1, flake8 7.3.0, djhtml 3.0.10, ripsecrets v0.1.11) - Add CLAUDE.md with rule prohibiting real contact identifiers in code Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-03-06 22:38:06 +00:00
Mark Veidemanis	ff66bc9e1f	Implement 3 plans	2026-03-06 19:38:32 +00:00
Mark Veidemanis	49aaed5dec	Fix XMPP auth	2026-03-06 18:21:24 +00:00
Mark Veidemanis	8c091b1e6d	Increase platform abstraction cohesion	2026-03-06 17:47:58 +00:00
Mark Veidemanis	438e561da0	Harden security	2026-03-05 05:42:19 +00:00
Mark Veidemanis	06735bdfb1	Begin adding AI memory	2026-03-05 03:24:39 +00:00
Mark Veidemanis	f21abd6299	Add minimal sensitive-info hygiene plan and projection cause sample diagnostics	2026-03-05 02:21:21 +00:00
Mark Veidemanis	2140c5facf	Lightweight containerized prosody tooling + moved auth scripts + xmpp reconnect/auth stabilization	2026-03-05 02:18:12 +00:00
Mark Veidemanis	0718a06c19	Implement plans	2026-03-04 02:19:22 +00:00
Mark Veidemanis	34ee49410d	Use plain Bulma themes when possible	2026-03-03 18:02:04 +00:00
Mark Veidemanis	8e0be6ca89	Theme capsules more uniformly	2026-03-03 17:58:22 +00:00
Mark Veidemanis	9646931181	Fix Signal emoji sync	2026-03-03 17:50:46 +00:00
Mark Veidemanis	506ea8a3b8	Remove raw IDs from some places	2026-03-03 17:45:50 +00:00
Mark Veidemanis	8ea2afb259	Add authors to tasks	2026-03-03 17:43:12 +00:00
Mark Veidemanis	18351abb00	Implement workspace history reconciliation	2026-03-03 17:35:45 +00:00
Mark Veidemanis	2898d9e832	Improve tasks and backdate insights	2026-03-03 17:21:06 +00:00
Mark Veidemanis	9c14e51b43	Implement executing tasks	2026-03-03 16:41:28 +00:00
Mark Veidemanis	d6bd56dace	Fix Signal messages and replies	2026-03-03 15:51:58 +00:00
Mark Veidemanis	56c620473f	Improve tasks and align page elements	2026-03-02 13:33:04 +00:00
Mark Veidemanis	e1de6d016d	Implement tasks	2026-03-02 12:45:24 +00:00
Mark Veidemanis	6986c1b5ab	Allow changing phrasing	2026-03-02 03:35:12 +00:00
Mark Veidemanis	00588ed1b8	Allow sharing conversations	2026-03-02 03:23:52 +00:00
Mark Veidemanis	b94219fc5b	Improve search	2026-03-02 02:26:25 +00:00
Mark Veidemanis	a9f5f3f75d	Fix business plans	2026-03-02 00:23:27 +00:00
Mark Veidemanis	b3e183eb0a	Implement business plans	2026-03-02 00:00:53 +00:00
Mark Veidemanis	d22924f6aa	Fix containerised restarts	2026-03-01 17:27:48 +00:00
Mark Veidemanis	4521755344	Fix Signal receiving	2026-02-22 18:05:26 +00:00
Mark Veidemanis	0f36b2dde7	Fix Signal compose live updates and self-chat direction	2026-02-22 11:03:08 +00:00
Mark Veidemanis	65cd647f01	Allow linking chats	2026-02-19 17:13:34 +00:00