Add filter

core/messaging/replies.py

@@ -7,6 +7,8 @@ import asyncio
 from django.utils import timezone
 import random
 
+log = logs.get_logger("replies")
+
 def should_reply(
         reply_to_self,
         reply_to_others,
@@ -24,7 +26,65 @@ def should_reply(
 
     return reply
 
-def generate_reply_prompt(msg: dict, person: Person, manip: Manipulation, chat_history: str):
+def generate_mutate_reply_prompt(msg: dict, person: Person, manip: Manipulation, chat_history: str, mutate: bool = False):
+    """
+    Strictly rewrites the message in the persona’s tone and style
+    while keeping the original meaning. No added explanations.
+    """
+
+    persona = manip.persona
+
+    # 🔹 **Strict Rules to Prevent AI from Adding Commentary**
+    strict_rules = (
+        "- **DO NOT add explanations, comments, or meta-thoughts**.\n"
+        "- **DO NOT return multiple responses—return ONLY the rewritten message**.\n"
+        "- **DO NOT change the meaning, intent, or facts in the message**.\n"
+        "- **DO NOT soften insults unless the persona naturally would**.\n"
+        "- **DO NOT reframe as a question or suggestion—this is NOT a conversation**.\n"
+        "- **Rewrite as if the original sender wrote it this way, without extra commentary**.\n"
+        "- **Start immediately with the rewritten message—NO preface, intro, or context.**\n"
+    )
+
+    # 🔹 **What the AI SHOULD do**
+    transformation_guidelines = (
+        "- **Rewrite the message in the persona’s unique tone and style**.\n"
+        "- **If the message is rude or harsh, reword it to match the persona’s confidence, cleverness, or wit**.\n"
+        "- **If the persona is sarcastic, teasing, or flirty, maintain that energy**.\n"
+        "- **Ensure the message feels natural, as if originally written that way**.\n"
+        "- **Preserve original sentence structure as much as possible, adjusting only for flow.**\n"
+    )
+
+    system_message = (
+        "You are a text rewriter. Your task is to transform messages into a given persona’s unique style, "
+        "while keeping the original meaning intact.\n\n"
+        "### Persona Profile ###\n"
+        f"- **Tone:** {persona.tone} | **Humor:** {persona.humor_style}\n"
+        f"- **Core Values:** {persona.core_values}\n"
+        f"- **Communication Style:** {persona.communication_style}\n"
+        f"- **Flirting Style:** {persona.flirting_style}\n"
+        f"- **Likes:** {persona.likes} | **Dislikes:** {persona.dislikes}\n"
+        f"- **Response Tactics:** {persona.response_tactics}\n"
+        f"- **Persuasion Techniques:** {persona.persuasion_tactics}\n"
+        f"- **Boundaries:** {persona.boundaries} | **Adaptability:** {persona.adaptability}%\n\n"
+
+        "### STRICT RULES ###\n"
+        f"{strict_rules}\n\n"
+
+        "### TRANSFORMATION GUIDELINES ###\n"
+        f"{transformation_guidelines}\n\n"
+
+        "### Original Message ###\n"
+        f"{msg}\n\n"
+
+        "### Rewritten Message ###\n"
+        "(DO NOT include anything except the rewritten text. NO extra comments or formatting.)"
+    )
+
+    return [{"role": "system", "content": system_message}]
+
+
+
+def generate_reply_prompt(msg: dict, person: Person, manip: Manipulation, chat_history: str, mutate: bool = False):
     """
     Generate a structured prompt using the attributes of the provided Person and Manipulation models.
     """
@@ -32,9 +92,23 @@ def generate_reply_prompt(msg: dict, person: Person, manip: Manipulation, chat_h
     now = timezone.now()
     persona = manip.persona
 
+    # 🔹 Define system instructions for filtering messages
+    filter_rules = (
+        "- **Strict Filtering**: If the message includes topics that are in dislikes or boundaries, "
+        "modify or reject them completely to match the persona's values.\n"
+        "- **Rephrase Sensitively**: If the message contains something the persona dislikes but is not "
+        "a strict boundary, soften it or remove the negative elements.\n"
+        "- **Enforce Persona Style**: Modify the tone, humor, and engagement level based on the persona's "
+        "communication style.\n"
+        "- **Never Ask for Clarifications**: Always modify the message directly without requesting clarification.\n"
+        "- **Return Only the Modified Message**: Do not add any explanations or metadata.\n"
+    )
+
     system_message = (
         "You are my digital persona, responding on my behalf while embodying my personality, preferences, and unique style.\n\n"
-        
+        "You must strictly apply the following persona-based filtering rules when modifying the message:\n\n"
+        f"{filter_rules}\n\n"
+
         "### Persona Profile ###\n"
         f"- **MBTI:** {persona.mbti} ({persona.mbti_identity} balance)\n"
         f"- **Tone:** {persona.tone} | **Humor:** {persona.humor_style}\n"
@@ -57,19 +131,19 @@ def generate_reply_prompt(msg: dict, person: Person, manip: Manipulation, chat_h
 
         "### Conversation Context ###\n"
         f"{chat_history if chat_history else 'No prior chat history.'}\n\n"
-
-        "### Response Guidelines ###\n"
-        "- **Engagement**: Keep responses engaging, with a balance of wit, depth, and confidence.\n"
-        "- **Flirting**: Be direct, playful, and, when appropriate, subtly provocative—without hesitation.\n"
-        "- **Pauses**: Use double newlines (`\\n\\n`) to pause where it enhances realism.\n"
-        "- **Flow Awareness**: Maintain continuity, avoid redundancy, and adjust response length based on interaction.\n"
     )
 
-    user_message = f"[{msg['timestamp']}] <{person.name}> {msg['text']}"
+    if not mutate:
+        user_message = f"[{msg['timestamp']}] <{person.name}> {msg['text']}"
+        log.info(f"User message: {user_message}")
 
-    return [
-        {"role": "system", "content": system_message},
-        {"role": "user", "content": user_message},
-    ]
-
-    
+        return [
+            {"role": "system", "content": system_message},
+            {"role": "user", "content": user_message},
+        ]
+    else:
+        user_message = f"Message to amend: {msg}"
+        return [
+            {"role": "system", "content": system_message},
+            {"role": "user", "content": user_message},
+        ]