# 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.