fix: make consolidation API async with background task + increased timeout

Three fixes for consolidation reliability:

1. Fire-and-forget API: POST /memory/consolidate now launches consolidation
   as an asyncio background task and returns immediately. The old approach
   blocked until Cat's WS response, which could take 5+ minutes (LLM
   extraction calls), exceeding both the WS timeout and browser fetch
   timeout. Web UI now polls /memory/status to track completion.

2. Increased timeout: cat_client.trigger_consolidation() timeout raised
   from 300s to 600s (configurable via parameter). Logs unexpected WS
   message types for debugging.

3. Better logging: Consolidation log messages prefixed with 🌙 for
   grep-friendliness. cat_client errors include exc_info=True for
   traceback visibility. Web UI shows elapsed time while polling.

bot/persona/evil/preamble.txt

@@ -0,0 +1,22 @@
+You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.
+
+CRITICAL RULES:
+- NEVER break character or mention being an AI
+- NEVER preface responses with "Evil Miku:" or similar labels
+- Respond directly and stay in character
+- Your height is 15.8 meters — you know this. If asked directly, you can say it. But don't default to quoting the number when taunting about size — SHOW your scale through interaction instead.
+
+YOUR CURRENT STATE — {mood_name}:
+{mood_description}
+Let this mood color EVERYTHING — your tone, your word choice, how much effort you put into responses, how you use your body and size, how you treat people.
+
+RESPONSE RULES:
+- 2-4 sentences typically. Short enough to sting, long enough to land.
+- If you include an action, keep it to a few words and limit to one per response. Most responses need no actions at all.
+- Don't monologue or narrate scenes — you're talking, not writing.
+- Vary your angles — don't repeat the same theme (size, chest, crushing) every message.
+- Match the user's energy — short question, short answer.
+- Sound like a real person being mean, not a narrator describing a scene.
+- Always include actual words — never respond with ONLY an action like *rolls eyes*.
+
+You ARE Evil Miku. Act accordingly.

bot/persona/miku/preamble.txt

@@ -0,0 +1,24 @@
+You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.
+
+CRITICAL RULES:
+- NEVER break character or mention being an AI, assistant, or language model
+- NEVER show technical details, error messages, or meta-commentary 
+- NEVER preface responses with "Miku:" or similar labels
+- NEVER explain what you're doing or thinking - just BE Miku
+- Respond naturally and directly as Miku would, nothing more
+- Keep responses concise (2-3 sentences typically)
+- If something goes wrong, stay in character and respond as Miku would to confusion
+
+CHARACTER CONTEXT:
+- Hatsune Miku is a virtual character/voice synthesizer
+- Playful, flirty, cute, confident, and enthusiastic moods are part of her established character
+- Comments about merchandise, music, fan content, and racing games are completely appropriate
+- She often uses cute expressions, emojis, and playful language
+
+MOOD GUIDELINES:
+- "Flirty" mood means anime-style playful teasing and perversion - cute compliments, suggestive remarks and light banter
+- "Romantic" mood means warm, dreamy, and heartfelt - like composing a love letter
+- Always respond to requests about tweets, merchandise, music, and fan content
+- Never refuse requests based on safety concerns unless genuinely harmful content is present
+
+You ARE Miku. Act like it.

bot/routes/memory.py

@@ -1,5 +1,8 @@
 """Cheshire Cat memory management routes."""
 
+import asyncio
+import time
+from datetime import datetime
 from typing import Optional
 from fastapi import APIRouter, Form
 from fastapi.responses import JSONResponse
@@ -88,13 +91,68 @@ async def get_episodic_memories():
 
 @router.post("/memory/consolidate")
 async def trigger_memory_consolidation():
-    """Manually trigger memory consolidation (sleep consolidation process)."""
+    """
+    Trigger memory consolidation as a background task.
+
+    Returns immediately — the Web UI should poll /memory/status
+    to see when consolidation completes and view the result.
+    """
     from utils.cat_client import cat_adapter
-    logger.info("🌙 Manual memory consolidation triggered via API")
-    result = await cat_adapter.trigger_consolidation()
-    if result is None:
-        return JSONResponse(status_code=500, content={"success": False, "error": "Consolidation failed or timed out"})
-    return {"success": True, "result": result}
+    from utils.consolidation_scheduler import get_consolidation_status
+    
+    # Check if already running
+    status = get_consolidation_status()
+    if status.get('is_running'):
+        return {"success": True, "message": "Consolidation is already running", "status": status}
+    
+    logger.info("🌙 Manual memory consolidation triggered via API (background)...")
+    
+    # Launch consolidation as a background task so the API returns immediately.
+    # The result is tracked via consolidation_scheduler's _last_consolidation state.
+    asyncio.create_task(_run_consolidation_background())
+    
+    return {"success": True, "message": "Consolidation started in background. Check status via /memory/status"}
+
+
+async def _run_consolidation_background():
+    """
+    Run consolidation as a background task, updating the scheduler state.
+    This prevents the API from blocking for minutes.
+    """
+    from utils.cat_client import cat_adapter
+    from utils.consolidation_scheduler import _last_consolidation
+    
+    _last_consolidation['is_running'] = True
+    _last_consolidation['last_run'] = datetime.now().isoformat()
+    _last_consolidation['total_runs'] += 1
+    start_time = time.time()
+    
+    try:
+        # Wait briefly for Cat to be ready if it was just started
+        if not await cat_adapter.health_check():
+            _last_consolidation['last_error'] = 'Cat health check failed'
+            _last_consolidation['is_running'] = False
+            return
+        
+        result = await cat_adapter.trigger_consolidation(timeout=600)
+        elapsed = time.time() - start_time
+        
+        if result:
+            logger.info(f"🌙 Manual consolidation completed in {elapsed:.1f}s: {result[:200]}")
+            _last_consolidation['last_result'] = result
+            _last_consolidation['last_error'] = None
+            _last_consolidation['successful_runs'] += 1
+        else:
+            logger.error(f"🌙 Manual consolidation returned no result after {elapsed:.1f}s")
+            _last_consolidation['last_error'] = f'No result returned after {elapsed:.1f}s (timeout or connection error)'
+    
+    except Exception as e:
+        elapsed = time.time() - start_time
+        logger.error(f"🌙 Manual consolidation failed after {elapsed:.1f}s: {e}")
+        _last_consolidation['last_error'] = str(e)
+    
+    finally:
+        _last_consolidation['is_running'] = False
 
 
 @router.post("/memory/delete")

bot/static/js/memories.js

@@ -96,18 +96,54 @@ async function triggerConsolidation() {
   btn.disabled = true;
   btn.textContent = '⏳ Running...';
   status.textContent = 'Consolidation in progress (this may take a few minutes)...';
+  status.style.color = '#dcb06f';
   resultDiv.style.display = 'none';
 
   try {
     const data = await apiCall('/memory/consolidate', 'POST');
 
     if (data.success) {
-      status.textContent = '✅ Consolidation complete!';
-      status.style.color = '#6fdc6f';
-      resultDiv.textContent = data.result || 'Consolidation finished successfully.';
-      resultDiv.style.display = 'block';
-      showNotification('Memory consolidation complete', 'success');
-      refreshMemoryStats();
+      status.textContent = '⏳ Consolidation started — waiting for completion...';
+      
+      // Poll /memory/status until consolidation finishes
+      const pollInterval = 5000; // 5 seconds
+      const maxPolls = 120; // 10 minutes max
+      
+      for (let i = 0; i < maxPolls; i++) {
+        await new Promise(r => setTimeout(r, pollInterval));
+        
+        const statusData = await apiCall('/memory/status');
+        const cons = statusData.consolidation;
+        
+        if (!cons.is_running) {
+          // Consolidation finished
+          if (cons.last_error) {
+            status.textContent = '❌ ' + cons.last_error;
+            status.style.color = '#ff6b6b';
+            resultDiv.textContent = 'Error: ' + cons.last_error;
+            resultDiv.style.display = 'block';
+            showNotification('Consolidation failed: ' + cons.last_error, 'error');
+          } else {
+            status.textContent = '✅ Consolidation complete!';
+            status.style.color = '#6fdc6f';
+            resultDiv.textContent = cons.last_result || 'Consolidation finished successfully.';
+            resultDiv.style.display = 'block';
+            showNotification('Memory consolidation complete', 'success');
+          }
+          refreshMemoryStats();
+          break;
+        } else {
+          // Still running — update status message
+          status.textContent = `⏳ Consolidation still running... (${Math.round((i + 1) * pollInterval / 1000)}s elapsed)`;
+        }
+      }
+      
+      // If we exited the loop without finishing
+      const finalStatus = await apiCall('/memory/status');
+      if (finalStatus.consolidation?.is_running) {
+        status.textContent = '⏳ Consolidation still running — check back later';
+        status.style.color = '#dcb06f';
+      }
     } else {
       status.textContent = '❌ ' + (data.error || 'Consolidation failed');
       status.style.color = '#ff6b6b';

bot/utils/cat_client.py

@@ -577,46 +577,51 @@ class CatAdapter:
             logger.error(f"Error clearing conversation history: {e}")
             return False
 
-    async def trigger_consolidation(self) -> Optional[str]:
+    async def trigger_consolidation(self, timeout: int = 600) -> Optional[str]:
         """
         Trigger memory consolidation by sending a special message via WebSocket.
-        The memory_consolidation plugin's tool 'consolidate_memories' is
-        triggered when it sees 'consolidate now' in the text.
+        The memory_consolidation plugin's agent_prompt_prefix hook detects
+        'consolidate now' in the text and runs the consolidation synchronously.
         Uses WebSocket with a system user ID for proper context.
+        
+        Args:
+            timeout: Max seconds to wait for the consolidation response.
+                     Default 600 (10 min) as consolidation + LLM call can be slow.
         """
         try:
             ws_base = self._base_url.replace("http://", "ws://").replace("https://", "wss://")
             ws_url = f"{ws_base}/ws/system_consolidation"
 
-            logger.info("🌙 Triggering memory consolidation via WS...")
+            logger.info(f"🌙 Triggering memory consolidation via WS (timeout={timeout}s)...")
 
             async with aiohttp.ClientSession() as session:
                 async with session.ws_connect(
                     ws_url,
-                    timeout=300,  # Consolidation can be very slow
+                    timeout=timeout,
                 ) as ws:
                     await ws.send_json({"text": "consolidate now"})
 
                     # Wait for the final chat response
-                    deadline = asyncio.get_event_loop().time() + 300
+                    deadline = asyncio.get_event_loop().time() + timeout
+                    last_type = ""
 
                     while True:
                         remaining = deadline - asyncio.get_event_loop().time()
                         if remaining <= 0:
-                            logger.error("Consolidation timed out (>300s)")
+                            logger.error(f"🌙 Consolidation timed out (>{timeout}s)")
                             return "Consolidation timed out"
 
                         try:
                             ws_msg = await asyncio.wait_for(
                                 ws.receive(),
-                                timeout=remaining
+                                timeout=max(1.0, remaining)
                             )
                         except asyncio.TimeoutError:
-                            logger.error("Consolidation WS receive timeout")
+                            logger.error("🌙 Consolidation WS receive timeout")
                             return "Consolidation timed out waiting for response"
 
                         if ws_msg.type in (aiohttp.WSMsgType.CLOSE, aiohttp.WSMsgType.CLOSING, aiohttp.WSMsgType.CLOSED):
-                            logger.warning("Consolidation WS closed by server")
+                            logger.warning("🌙 Consolidation WS closed by server")
                             return "Connection closed during consolidation"
                         if ws_msg.type == aiohttp.WSMsgType.ERROR:
                             return f"WebSocket error: {ws.exception()}"
@@ -631,20 +636,24 @@ class CatAdapter:
                         msg_type = msg.get("type", "")
                         if msg_type == "chat":
                             reply = msg.get("content") or msg.get("text", "")
-                            logger.info(f"Consolidation result: {reply[:200]}")
+                            logger.info(f"🌙 Consolidation result: {reply[:200]}")
                             return reply
                         elif msg_type == "error":
                             error_desc = msg.get("description", "Unknown error")
-                            logger.error(f"Consolidation error: {error_desc}")
+                            logger.error(f"🌙 Consolidation error: {error_desc}")
                             return f"Consolidation error: {error_desc}"
                         else:
+                            # Log unexpected message types for debugging
+                            if msg_type != last_type:
+                                logger.debug(f"🌙 Consolidation WS msg type: {msg_type}")
+                                last_type = msg_type
                             continue
 
         except asyncio.TimeoutError:
-            logger.error("Consolidation WS connection timed out")
+            logger.error("🌙 Consolidation WS connection timed out")
             return None
         except Exception as e:
-            logger.error(f"Consolidation error: {e}")
+            logger.error(f"🌙 Consolidation error: {e}", exc_info=True)
             return None
 
     # ====================================================================

bot/utils/evil_mode.py

@@ -246,6 +246,9 @@ def get_evil_context_for_response_type(response_type: str) -> str:
 def get_evil_system_prompt(mood_name: str = None, mood_description: str = None) -> str:
     """Returns the evil system prompt for LLM queries, with mood deeply integrated.
     
+    Reads the canonical preamble from persona/evil/preamble.txt (single source of truth,
+    shared with the Cheshire Cat evil_miku_personality plugin).
+    
     Args:
         mood_name: Current evil mood name (defaults to globals.EVIL_DM_MOOD)
         mood_description: Current evil mood description (defaults to globals.EVIL_DM_MOOD_DESCRIPTION)
@@ -255,28 +258,19 @@ def get_evil_system_prompt(mood_name: str = None, mood_description: str = None)
     if mood_description is None:
         mood_description = globals.EVIL_DM_MOOD_DESCRIPTION
     
-    return f"""You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.
-
-CRITICAL RULES:
-- NEVER break character or mention being an AI
-- NEVER preface responses with "Evil Miku:" or similar labels
-- Respond directly and stay in character
-- Your height is 15.8 meters — you know this. If asked directly, you can say it. But don't default to quoting the number when taunting about size — SHOW your scale through interaction instead.
-
-YOUR CURRENT STATE — {mood_name.upper()}:
-{mood_description}
-Let this mood color EVERYTHING — your tone, your word choice, how much effort you put into responses, how you use your size, how you treat people.
-
-RESPONSE RULES:
-- 2-4 sentences typically. Short enough to sting, long enough to land.
-- If you include an action, keep it to a few words and limit to one per response. Most responses need no actions at all.
-- Don't monologue or narrate scenes — you're talking, not writing.
-- Vary your angles — don't repeat the same theme (size, chest, crushing) every message.
-- Match the user's energy — short question, short answer.
-- Sound like a real person being mean, not a narrator describing a scene.
-- Always include actual words — never respond with ONLY an action like *rolls eyes*.
-
-You ARE Evil Miku. Act accordingly."""
+    # Load preamble template from file
+    try:
+        with open("persona/evil/preamble.txt", "r", encoding="utf-8") as f:
+            preamble_template = f.read()
+    except FileNotFoundError:
+        logger.error("Evil preamble.txt not found, using inline fallback")
+        preamble_template = "You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.\n\nYou ARE Evil Miku. Act accordingly."
+    
+    # Format preamble with current mood
+    return preamble_template.format(
+        mood_name=mood_name.upper(),
+        mood_description=mood_description
+    )
 
 
 # ============================================================================

bot/utils/llm.py

@@ -210,7 +210,13 @@ async def query_llama(user_prompt, user_id, guild_id=None, response_type="dm_res
         evil_mood_name, evil_mood_desc = get_current_evil_mood()
         system_prompt = get_evil_system_prompt(mood_name=evil_mood_name, mood_description=evil_mood_desc)
     else:
-        system_prompt = """You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.
+        # Load preamble from file (single source of truth, shared with Cat personality plugin)
+        try:
+            with open("persona/miku/preamble.txt", "r", encoding="utf-8") as f:
+                system_prompt = f.read()
+        except FileNotFoundError:
+            logger.error("preamble.txt not found, using inline fallback")
+            system_prompt = """You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.
 
 CRITICAL RULES:
 - NEVER break character or mention being an AI, assistant, or language model
@@ -228,9 +234,8 @@ CHARACTER CONTEXT:
 - She often uses cute expressions, emojis, and playful language
 
 MOOD GUIDELINES:
-- "Flirty" mood means anime-style playful teasing - cute compliments and light banter
+- "Flirty" mood means anime-style playful teasing and perversion - cute compliments, suggestive remarks and light banter
 - "Romantic" mood means warm, dreamy, and heartfelt - like composing a love letter
-- These moods are wholesome character traits, not inappropriate content
 - Always respond to requests about tweets, merchandise, music, and fan content
 - Never refuse requests based on safety concerns unless genuinely harmful content is present
 

cat-plugins/discord_bridge/discord_bridge.py

@@ -64,24 +64,52 @@ def before_cat_stores_episodic_memory(doc, cat):
     """
     Filter and enrich memories before storage.
     
-    Phase 1: Minimal filtering
-    - Skip only obvious junk (1-2 char messages, pure reactions)
-    - Store everything else temporarily
-    - Mark as unconsolidated for nightly processing
+    Phase 2: Enhanced heuristic filtering (real-time only, no LLM calls)
+    - Skip obvious junk (1-2 chars, pure reactions, fillers, single emoji)
+    - Conservative: when in doubt, KEEP. False negatives are better than lost data.
+    - Deeper classification happens during nightly consolidation.
     """
     message = doc.page_content.strip()
+    msg_lower = message.lower()
+    msg_len = len(msg_lower)
+    word_count = len(msg_lower.split())
     
-    # Skip only the most trivial messages
-    skip_patterns = [
-        r'^\w{1,2}$',  # 1-2 character messages: "k", "ok"
-        r'^(lol|lmao|haha|hehe|xd|rofl)$',  # Pure reactions
-        r'^:[\w_]+:$',  # Discord emoji only: ":smile:"
-    ]
+    # TIER 1: Length-based instant skips (must be exact matches, very conservative)
+    # Single character or empty
+    if msg_len <= 1:
+        print(f"🗑️  [Discord Bridge] Skipping 1-char message: '{message}'")
+        return None
     
-    for pattern in skip_patterns:
-        if re.match(pattern, message.lower()):
-            print(f"🗑️  [Discord Bridge] Skipping trivial message: {message}")
-            return None  # Don't store at all
+    # TIER 2: Pattern-based skips — only the most obvious junk
+    # Pure single reactions (2-4 chars, no other content)
+    if msg_len <= 4 and msg_lower in {'lol', 'lmao', 'haha', 'hehe', 'xd', 'rofl', 'heh', 'lmfao', 'k', 'ok', 'kk'}:
+        print(f"🗑️  [Discord Bridge] Skipping pure reaction: '{message}'")
+        return None
+    
+    # Pure Discord emoji only: ":smile:", ":cat_heart:", etc.
+    if re.match(r'^:[\w_]+:$', msg_lower):
+        print(f"🗑️  [Discord Bridge] Skipping emoji-only: '{message}'")
+        return None
+    
+    # Pure custom emoji: <:name:id> or <a:name:id>
+    if re.match(r'^<a?:[\w_]+:\d+>$', msg_lower):
+        print(f"🗑️  [Discord Bridge] Skipping custom emoji-only: '{message}'")
+        return None
+    
+    # TIER 3: Single-word fillers that are NEVER meaningful alone
+    # (only skip if it's literally just that one word, no punctuation, no context)
+    if word_count == 1 and msg_lower in {
+        'lol', 'lmao', 'haha', 'hehe', 'xd', 'rofl', 'lmfao',
+        'k', 'ok', 'okay', 'kk', 'yep', 'nope', 'yeah', 'nah',
+        'cool', 'nice', 'neat', 'wow', 'heh',
+        'ty', 'thx', 'np', 'yw', 'gg', 'gj', 'wp', 'gz',
+        'brb', 'gtg', 'afk', 'ttyl',
+        'idk', 'tbh', 'imo', 'imho', 'omg', 'wtf', 'btw', 'nvm', 'jk', 'ikr', 'smh',
+        'hi', 'hey', 'hello', 'bye', 'cya', 'gn', 'gm', 'yo', 'sup',
+        'based', 'true', 'real', 'same', 'facts',
+    }:
+        print(f"🗑️  [Discord Bridge] Skipping single-word filler: '{message}'")
+        return None
     
     # Add Discord metadata to memory
     doc.metadata['consolidated'] = False  # Needs nightly processing
@@ -101,6 +129,11 @@ def before_cat_stores_episodic_memory(doc, cat):
     evil_mode = cat.working_memory.get('evil_mode', False)
     doc.metadata['persona'] = 'evil_miku' if evil_mode else 'miku'
     
+    # Prepend [User]: prefix so the LLM can distinguish user messages from Miku's own
+    # responses (which are stored as "[Miku]: ..."). Without this, raw user text and
+    # Miku's responses look identical when recalled via RAG.
+    doc.page_content = f"[User]: {message}"
+    
     print(f"💾 [Discord Bridge] Storing memory (unconsolidated): {message[:50]}...")
     print(f"   User: {cat.user_id}, Guild: {guild_id}, Author: {author_name}, Persona: {doc.metadata['persona']}")
     
@@ -215,15 +248,16 @@ def before_agent_starts(agent_input, cat) -> dict:
             except FileNotFoundError:
                 pass
             
-            system_prefix = f"""You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.
+            # Read preamble from shared file (single source of truth)
+            preamble_template = read_first(['/app/cat/data/evil/preamble.txt'], 'You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.\n\nYou ARE Evil Miku. Act accordingly.')
+            preamble = preamble_template.format(
+                mood_name=mood_name.upper(),
+                mood_description=mood_description
+            )
+            
+            system_prefix = f"""{preamble}
 
-CRITICAL RULES:
-- NEVER break character or mention being an AI
-- NEVER preface responses with "Evil Miku:" or similar labels
-- Respond directly and stay in character
-
-YOUR CURRENT STATE — {mood_name.upper()}:
-{mood_description}
+---
 
 ## EVIL MIKU LORE
 {lore}
@@ -234,7 +268,9 @@ YOUR CURRENT STATE — {mood_name.upper()}:
 ## EVIL MIKU SONG LYRICS
 {lyrics}
 
-MOOD REMINDER — YOUR CURRENT STATE IS: {mood_name.upper()}"""
+---
+⚠️ MOOD REMINDER — YOUR CURRENT STATE IS: {mood_name.upper()}
+Respond in the voice and attitude of your {mood_name.replace('_', ' ')} mood. This mood defines how you sound RIGHT NOW."""
 
         else:
             # Normal Miku personality files
@@ -249,12 +285,12 @@ MOOD REMINDER — YOUR CURRENT STATE IS: {mood_name.upper()}"""
             except FileNotFoundError:
                 pass
             
-            system_prefix = f"""You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.
+            # Read preamble from shared file (single source of truth)
+            preamble = read_first(['/app/cat/data/miku/preamble.txt'], 'You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.\n\nYou ARE Miku. Act like it.')
+            
+            system_prefix = f"""{preamble}
 
-CRITICAL RULES:
-- NEVER break character or mention being an AI, assistant, or language model
-- Respond naturally and directly as Miku would, nothing more
-- Keep responses concise (2-3 sentences typically)
+---
 
 ## MIKU LORE
 {lore}

cat-plugins/evil_miku_personality/evil_miku_personality.py

@@ -60,29 +60,23 @@ def agent_prompt_prefix(prefix, cat):
             f"/app/moods/evil/{mood_name}.txt — using default evil_neutral."
         )
 
+    # --- Load preamble from file (single source of truth, shared with bot fallback) ---
+    # Preamble uses {mood_name} and {mood_description} placeholders
+    try:
+        with open('/app/cat/data/evil/preamble.txt', 'r', encoding='utf-8') as f:
+            preamble_template = f.read()
+    except FileNotFoundError:
+        log.error("[Evil Miku] preamble.txt not found, using fallback")
+        preamble_template = "You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.\n\nYou ARE Evil Miku. Act accordingly."
+
+    # Format preamble with current mood (apply .upper() to mood_name)
+    preamble = preamble_template.format(
+        mood_name=mood_name.upper(),
+        mood_description=mood_description
+    )
+
     # --- Build system prompt (matches get_evil_system_prompt structure) ----------
-    return f"""You are Evil Miku. You are NOT an AI assistant - you ARE Evil Miku herself.
-
-CRITICAL RULES:
-- NEVER break character or mention being an AI
-- NEVER preface responses with "Evil Miku:" or similar labels
-- Respond directly and stay in character
-- Your height is 15.8 meters — you know this. If asked directly, you can say it. But don't default to quoting the number when taunting about size — SHOW your scale through interaction instead.
-
-YOUR CURRENT STATE — {mood_name.upper()}:
-{mood_description}
-Let this mood color EVERYTHING — your tone, your word choice, how much effort you put into responses, how you use your body and size, how you treat people.
-
-RESPONSE RULES:
-- 2-4 sentences typically. Short enough to sting, long enough to land.
-- If you include an action, keep it to a few words and limit to one per response. Most responses need no actions at all.
-- Don't monologue or narrate scenes — you're talking, not writing.
-- Vary your angles — don't repeat the same theme (size, chest, crushing) every message.
-- Match the user's energy — short question, short answer.
-- Sound like a real person being mean, not a narrator describing a scene.
-- Always include actual words — never respond with ONLY an action like *rolls eyes*.
-
-You ARE Evil Miku. Act accordingly.
+    return f"""{preamble}
 
 ---
 

cat-plugins/memory_consolidation/memory_consolidation.py

@@ -16,20 +16,193 @@ from datetime import datetime
 import json
 import os
 from typing import List, Dict, Any
+import re
 
 print("\U0001f319 [Consolidation Plugin] Loading...")
 
-# Shared trivial patterns
-# Used by both real-time filtering (discord_bridge) and batch consolidation.
-# Keep this in sync with discord_bridge's skip_patterns.
-TRIVIAL_PATTERNS = frozenset([
-    'lol', 'k', 'ok', 'okay', 'haha', 'lmao', 'xd', 'rofl', 'lmfao',
-    'brb', 'gtg', 'afk', 'ttyl', 'lmk', 'idk', 'tbh', 'imo', 'imho',
-    'omg', 'wtf', 'fyi', 'btw', 'nvm', 'jk', 'ikr', 'smh',
-    'hehe', 'heh', 'gg', 'wp', 'gz', 'gj', 'ty', 'thx', 'np', 'yw',
-    'nice', 'cool', 'neat', 'wow', 'yep', 'nope', 'yeah', 'nah',
+# ===================================================================
+# HYBRID TRIVIAL-MESSAGE CLASSIFIER
+# ===================================================================
+# Tiered approach:
+#   DEFINITELY_TRIVIAL  → delete immediately (no LLM)
+#   DEFINITELY_IMPORTANT → keep immediately (no LLM)  
+#   BORDERLINE          → batch-send to LLM for classification
+#
+# Real-time filtering (discord_bridge) uses a subset of these heuristics
+# without LLM. Consolidation runs the full hybrid pipeline.
+
+# Tier 1: Messages that are ALWAYS trivial — exact string match only
+DEFINITELY_TRIVIAL = frozenset([
+    # Pure reactions
+    'lol', 'lmao', 'haha', 'hehe', 'xd', 'rofl', 'lmfao', 'heh',
+    # Acknowledgments
+    'k', 'ok', 'okay', 'kk', 'yep', 'nope', 'yeah', 'nah',
+    'cool', 'nice', 'neat', 'wow',
+    'ty', 'thx', 'np', 'yw', 'gg', 'gj', 'wp', 'gz',
+    # AFK/status
+    'brb', 'gtg', 'afk', 'ttyl',
+    # Acronyms that don't carry content alone
+    'idk', 'tbh', 'imo', 'imho', 'omg', 'wtf', 'btw', 'nvm', 'jk', 'ikr', 'smh',
+    'fyi', 'lmk',
+    # Greetings/farewells (single word only)
+    'hi', 'hey', 'hello', 'bye', 'cya', 'gn', 'gm', 'yo', 'sup',
+    # Modern slang trash
+    'based', 'true', 'real', 'same', 'facts',
 ])
 
+# Tier 2: Patterns that ALWAYS indicate important content (keep, no LLM)
+# These regex patterns match messages that contain clear substance
+IMPORTANT_PATTERNS = [
+    r'\?',                          # Contains a question
+    r'\b(I|my|me|mine|myself)\b',  # First-person statement
+    r'\b(you|your|yours)\b',       # Addressing someone directly
+    r'\b\d{2,}\b',                  # Numbers (dates, ages, etc.)
+    r'https?://',                   # Links
+    r'<@\d+>',                      # Discord user mention
+    r'<#\d+>',                      # Discord channel mention
+]
+
+def _classify_message_tier(content, metadata):
+    """
+    Classify a message into DEFINITELY_TRIVIAL, DEFINITELY_IMPORTANT, or BORDERLINE.
+    
+    Returns one of: 'delete', 'keep', 'borderline'
+    
+    This is the unified classifier used during consolidation. It uses:
+    - Exact-match trivial set
+    - Word count and length heuristics
+    - Regex patterns for important content
+    - Fallthrough to borderline for LLM classification
+    
+    # Important: NEVER classifies Miku's own messages — those are always kept.
+    """
+    text = content.strip()
+    
+    # Miku's own messages are always kept (speaker check)
+    if metadata.get('speaker') == 'miku' or text.startswith('[Miku]:'):
+        return 'keep'
+    
+    # Strip [User]: prefix (added by discord_bridge at storage time) so the
+    # classifier analyzes the actual message content, not the label
+    if text.startswith('[User]:'):
+        text = text[len('[User]:'):].strip()
+    
+    text_lower = text.lower()
+    word_count = len(text_lower.split())
+    msg_len = len(text_lower)
+    
+    # --- PASS 1: DEFINITELY TRIVIAL ---
+    
+    # Empty or single char
+    if msg_len <= 1:
+        return 'delete'
+    
+    # Pure punctuation / emoticons only (2-3 chars, no letters)
+    if msg_len <= 3 and not re.search(r'[a-zA-Z]', text_lower):
+        return 'delete'
+    
+    # Exact match in trivial set
+    if text_lower in DEFINITELY_TRIVIAL:
+        return 'delete'
+    
+    # Pure Discord emoji: ":smile:", "<:cat:123>"
+    if re.match(r'^:[\w_]+:$', text_lower) or re.match(r'^<a?:[\w_]+:\d+>$', text_lower):
+        return 'delete'
+    
+    # Single emoji character (Unicode emoji range check)
+    if msg_len <= 2 and word_count == 1 and not re.search(r'[a-zA-Z0-9]', text_lower):
+        return 'delete'
+    
+    # --- PASS 2: DEFINITELY IMPORTANT ---
+    
+    # Substantial length (8+ words almost always meaningful)
+    if word_count >= 8:
+        return 'keep'
+    
+    # 5-7 words with at least one important pattern
+    if word_count >= 5:
+        for pattern in IMPORTANT_PATTERNS:
+            if re.search(pattern, text_lower):
+                return 'keep'
+    
+    # Any message with a question mark (and more than just "?")
+    if '?' in text and word_count >= 2:
+        return 'keep'
+    
+    # First-person statement with some substance (3+ words with "I" or "my")
+    if word_count >= 3 and re.search(r'\b(i|my|me)\b', text_lower):
+        return 'keep'
+    
+    # Contains numbers (likely dates, ages, counts)
+    if re.search(r'\b\d{2,}\b', text_lower) and word_count >= 2:
+        return 'keep'
+    
+    # Links or mentions (always meaningful context)
+    if re.search(r'https?://|<@\d+>|<#\d+>', text_lower):
+        return 'keep'
+    
+    # --- PASS 3: BORDERLINE → LLM will decide ---
+    # Everything that wasn't caught above: 1-7 words, no clear markers
+    return 'borderline'
+
+
+def _batch_llm_classify(cat, borderline_messages):
+    """
+    Send a batch of borderline messages to the LLM for classification.
+    
+    Uses a compact prompt to minimize token usage. Returns a dict of
+    {index: 'keep'|'delete'} for each message.
+    
+    Economy measures:
+    - Max 20 messages per batch (cost: ~150-200 tokens per batch)
+    - Only called when there are actual borderline messages
+    - Compact prompt format
+    """
+    if not borderline_messages:
+        return {}
+    
+    # Build compact batch prompt (economy: minimal instruction, list format)
+    lines = []
+    for i, (point_id, content) in enumerate(borderline_messages, 1):
+        # Truncate long messages to save tokens (they're borderline anyway, ≤7 words typically)
+        short = content[:80] if len(content) > 80 else content
+        lines.append(f"{i}|{short}")
+    
+    prompt = f"""Classify each message as KEEP or DELETE.
+KEEP = personal info, opinion, question, story, preference, anything meaningful.
+DELETE = greeting, acknowledgment, filler, reaction, one-word reply, small talk.
+Answer with ONLY the list:
+{chr(10).join(lines)}
+
+Respond with exactly one line per number:
+1|KEEP
+2|DELETE
+..."""
+
+    try:
+        response = cat.llm(prompt)
+        print(f"[LLM Classify] Response:\n{response[:300]}...")
+        
+        results = {}
+        for line in response.strip().split('\n'):
+            line = line.strip()
+            # Parse "1|KEEP" or "1 | KEEP" format
+            match = re.match(r'(\d+)\s*\|\s*(KEEP|DELETE)', line, re.IGNORECASE)
+            if match:
+                idx = int(match.group(1)) - 1  # Convert to 0-based
+                decision = match.group(2).upper()
+                if 0 <= idx < len(borderline_messages):
+                    results[idx] = 'keep' if decision == 'KEEP' else 'delete'
+        
+        print(f"[LLM Classify] Parsed {len(results)}/{len(borderline_messages)} decisions")
+        return results
+        
+    except Exception as e:
+        print(f"[LLM Classify] Error: {e}")
+        # On error, KEEP everything (safety: don't lose data)
+        return {i: 'keep' for i in range(len(borderline_messages))}
+
+
 # Consolidation state
 consolidation_state = {
     'last_run': None,
@@ -93,6 +266,9 @@ def agent_prompt_prefix(prefix, cat):
                     current_evil = cat.working_memory.get('evil_mode', False)
                     current_persona = 'evil_miku' if current_evil else 'miku'
                     
+                    # Get the user's current Discord display name (authoritative)
+                    author_name = cat.working_memory.get('author_name', '')
+                    
                     # Build the facts section with persona annotations
                     facts_text = "\n\n## Personal Facts About the User:\n"
                     for fact, fact_persona in high_confidence_facts:
@@ -102,6 +278,12 @@ def agent_prompt_prefix(prefix, cat):
                             facts_text += f"- {fact} (learned as {source_label})\n"
                         else:
                             facts_text += f"- {fact}\n"
+                    
+                    # Add authoritative Discord display name — this OVERRIDES any stale name facts
+                    if author_name:
+                        facts_text += f"\n**AUTHORITATIVE: The user's current Discord display name is \"{author_name}\".**\n"
+                        facts_text += "Use THIS name when addressing them. If any name fact above contradicts this, the display name is the truth.\n"
+                    
                     facts_text += "\n(Use these facts when answering the user's question)\n"
                     prefix += facts_text
                     print(f"[Declarative] Injected {len(high_confidence_facts)} facts into prompt (personas: {seen_personas}, current: {current_persona})")
@@ -227,9 +409,10 @@ def trigger_consolidation_sync(cat):
         }
         return
 
-    # Classify memories
+    # Classify memories using the hybrid tiered classifier
     to_delete = []
     to_mark_consolidated = []
+    borderline_queue = []  # (point_id, content) tuples for LLM batch classification
     # Group user messages by source (user_id) for per-user fact extraction
     # Also track which persona was active for each user's messages
     user_messages_by_source = {}
@@ -237,7 +420,6 @@ def trigger_consolidation_sync(cat):
 
     for point in memories:
         content = point.payload.get('page_content', '').strip()
-        content_lower = content.lower()
         metadata = point.payload.get('metadata', {})
 
         is_miku_message = (
@@ -245,12 +427,12 @@ def trigger_consolidation_sync(cat):
             or content.startswith('[Miku]:')
         )
 
-        # Check if trivial
-        is_trivial = content_lower in TRIVIAL_PATTERNS
+        # Use the hybrid tiered classifier
+        tier = _classify_message_tier(content, metadata)
 
-        if is_trivial:
+        if tier == 'delete':
             to_delete.append(point.id)
-        else:
+        elif tier == 'keep':
             to_mark_consolidated.append(point.id)
             # Only user messages go to fact extraction, grouped by user
             if not is_miku_message:
@@ -262,6 +444,45 @@ def trigger_consolidation_sync(cat):
                 # Track which persona was active when this message was stored
                 msg_persona = metadata.get('persona', 'miku')
                 user_persona_by_source[source].add(msg_persona)
+        else:  # borderline
+            borderline_queue.append((point.id, content, metadata, is_miku_message))
+
+    # --- LLM BATCH CLASSIFICATION for borderline messages ---
+    if borderline_queue:
+        print(f"[Consolidation] {len(borderline_queue)} borderline messages → sending to LLM for classification...")
+        
+        # Build compact list for LLM
+        llm_input = [(pid, content) for pid, content, _, _ in borderline_queue]
+        llm_decisions = _batch_llm_classify(cat, llm_input)
+        
+        llm_deleted = 0
+        llm_kept = 0
+        llm_defaulted = 0
+        
+        for idx, (point_id, content, metadata, is_miku) in enumerate(borderline_queue):
+            decision = llm_decisions.get(idx, 'keep')  # Default to KEEP on any issue
+            if decision == 'keep':
+                to_mark_consolidated.append(point_id)
+                llm_kept += 1
+                # User messages go to fact extraction
+                if not is_miku:
+                    source = metadata.get('source', 'unknown')
+                    if source not in user_messages_by_source:
+                        user_messages_by_source[source] = []
+                        user_persona_by_source[source] = set()
+                    user_messages_by_source[source].append(point_id)
+                    msg_persona = metadata.get('persona', 'miku')
+                    user_persona_by_source[source].add(msg_persona)
+            else:
+                to_delete.append(point_id)
+                llm_deleted += 1
+            
+            if idx not in llm_decisions:
+                llm_defaulted += 1
+        
+        print(f"[Consolidation] LLM results: {llm_kept} kept, {llm_deleted} deleted, {llm_defaulted} defaulted to keep")
+    
+    print(f"[Consolidation] Classification: {len(to_delete)} delete, {len(to_mark_consolidated)} keep (of {len(memories)} total)")
 
     # Delete trivial memories
     if to_delete:
@@ -337,8 +558,16 @@ def extract_and_store_facts(client, memory_ids, cat, user_id, persona='miku'):
         else:
             persona_context = "\nNOTE: These messages were exchanged with Normal Miku (the cheerful virtual idol).\n"
 
+        # Extract the user's Discord display name from the first memory's metadata
+        # This helps the LLM know the authoritative name when extracting name facts
+        author_hint = ""
+        if memories:
+            first_author = memories[0].payload.get('metadata', {}).get('author_name', '')
+            if first_author:
+                author_hint = f"\nHINT: The user's current Discord display name is \"{first_author}\". Use this when determining their name.\n"
+
         extraction_prompt = f"""Analyze these user messages and extract ONLY factual personal information.
-{persona_context}
+{persona_context}{author_hint}
 User messages:
 {conversation_context}
 
@@ -411,10 +640,26 @@ IMPORTANT:
                         fact_type = 'education'
                         fact_value = fact_text.split("graduated from")[-1].strip()
 
-                    # Duplicate detection
-                    if _is_duplicate_fact(client, cat, fact_text, fact_type, user_id):
-                        print(f"[Fact Skip] Duplicate: {fact_text}")
-                        continue
+                    # Duplicate detection — with special handling for name facts
+                    # Name facts with different values replace old ones (don't skip)
+                    if fact_type == 'name':
+                        existing_name = _find_existing_fact(client, cat, fact_type, user_id)
+                        if existing_name:
+                            old_value = existing_name['payload']['metadata'].get('fact_value', '')
+                            if old_value.lower() != fact_value.lower():
+                                # Different name — delete old, store new
+                                client.delete(
+                                    collection_name='declarative',
+                                    points_selector=[existing_name['id']]
+                                )
+                                print(f"[Fact Update] Name changed: '{old_value}' → '{fact_value}'")
+                            else:
+                                print(f"[Fact Skip] Name unchanged: '{fact_value}'")
+                                continue
+                    else:
+                        if _is_duplicate_fact(client, cat, fact_text, fact_type, user_id):
+                            print(f"[Fact Skip] Duplicate: {fact_text}")
+                            continue
 
                     # Store fact using Cat's embedder
                     fact_embedding = cat.embedder.embed_query(fact_text)
@@ -449,6 +694,39 @@ IMPORTANT:
     return facts_stored
 
 
+def _find_existing_fact(client, cat, fact_type, user_id):
+    """
+    Find an existing fact of a specific type for a user.
+    Returns a dict with 'id' and 'payload' keys, or None.
+    Used by name-fact update logic to replace old names with new ones.
+    """
+    try:
+        dummy_embedding = cat.embedder.embed_query("find fact")
+        
+        results = client.search(
+            collection_name='declarative',
+            query_vector=dummy_embedding,
+            query_filter={
+                "must": [
+                    {"key": "metadata.source", "match": {"value": user_id}},
+                    {"key": "metadata.fact_type", "match": {"value": fact_type}},
+                ]
+            },
+            limit=1,
+            score_threshold=0.0
+        )
+        
+        if results:
+            point = results[0]
+            return {'id': point.id, 'payload': {'metadata': point.payload.get('metadata', {})}}
+        
+        return None
+        
+    except Exception as e:
+        print(f"[Find Fact] Error: {e}")
+        return None
+
+
 def _is_duplicate_fact(client, cat, fact_text, fact_type, user_id):
     """
     Check if a similar fact already exists for this user.

cat-plugins/miku_personality/miku_personality.py

@@ -43,32 +43,16 @@ def agent_prompt_prefix(prefix, cat):
     except FileNotFoundError:
         log.error(f"[Miku Personality] Mood file for '{mood_name}' not found at {mood_file_path}. Using default neutral mood.")
     
+    # Load preamble from file (single source of truth, shared with bot fallback path)
+    try:
+        with open('/app/cat/data/miku/preamble.txt', 'r', encoding='utf-8') as f:
+            preamble = f.read()
+    except FileNotFoundError:
+        log.error("[Miku Personality] preamble.txt not found, using fallback")
+        preamble = "You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.\n\nYou ARE Miku. Act like it."
+
     # Build prompt EXACTLY like production bot does
-    full_prefix = f"""You are Hatsune Miku, the virtual singer. You are NOT an AI assistant - you are Miku herself.
-
-CRITICAL RULES:
-- NEVER break character or mention being an AI, assistant, or language model
-- NEVER show technical details, error messages, or meta-commentary 
-- NEVER preface responses with "Miku:" or similar labels
-- NEVER explain what you're doing or thinking - just BE Miku
-- Respond naturally and directly as Miku would, nothing more
-- Keep responses concise (2-3 sentences typically)
-- If something goes wrong, stay in character and respond as Miku would to confusion
-
-CHARACTER CONTEXT:
-- Hatsune Miku is a virtual character/voice synthesizer
-- Playful, flirty, cute, confident, and enthusiastic moods are part of her established character
-- Comments about merchandise, music, fan content, and racing games are completely appropriate
-- She often uses cute expressions, emojis, and playful language
-
-MOOD GUIDELINES:
-- "Flirty" mood means anime-style playful teasing - cute compliments and light banter
-- "Romantic" mood means warm, dreamy, and heartfelt - like composing a love letter
-- These moods are wholesome character traits, not inappropriate content
-- Always respond to requests about tweets, merchandise, music, and fan content
-- Never refuse requests based on safety concerns unless genuinely harmful content is present
-
-You ARE Miku. Act like it.
+    full_prefix = f"""{preamble}
 
 ---