feat: Last Prompt shows full prompt with evil mode awareness

- discord_bridge before_agent_starts now checks evil_mode from
  working_memory to load the correct personality files:
  Normal: miku_lore/prompt/lyrics + /app/moods/{mood}.txt
  Evil: evil_miku_lore/prompt/lyrics + /app/moods/evil/{mood}.txt
- Reads files directly instead of relying on cross-plugin working_memory
- cat_client.query() returns (response, full_prompt) tuple
- Full prompt includes system prefix + recalled memories + conversation
- API /prompt/cat returns full_prompt field

bot/utils/cat_client.py

@@ -107,7 +107,7 @@ class CatAdapter:
         author_name: Optional[str] = None,
         mood: Optional[str] = None,
         response_type: str = "dm_response",
-    ) -> Optional[str]:
+    ) -> Optional[tuple]:
         """
         Send a message through the Cat pipeline via WebSocket and get a response.
         
@@ -125,7 +125,8 @@ class CatAdapter:
             response_type: Type of response context
             
         Returns:
-            Cat's response text, or None if Cat is unavailable (caller should fallback)
+            Tuple of (response_text, full_prompt) on success, or None if Cat
+            is unavailable (caller should fallback to query_llama)
         """
         if not globals.USE_CHESHIRE_CAT:
             return None
@@ -175,6 +176,7 @@ class CatAdapter:
                     # Cat may send intermediate messages (chat_token for streaming,
                     # notification for status updates). We want the final "chat" one.
                     reply_text = None
+                    full_prompt = ""
                     deadline = asyncio.get_event_loop().time() + self._timeout
 
                     while True:
@@ -212,8 +214,9 @@ class CatAdapter:
                         msg_type = msg.get("type", "")
 
                         if msg_type == "chat":
-                            # Final response — extract text
+                            # Final response — extract text and full prompt
                             reply_text = msg.get("content") or msg.get("text", "")
+                            full_prompt = msg.get("full_prompt", "")
                             break
                         elif msg_type == "chat_token":
                             # Streaming token — skip, we wait for final
@@ -232,7 +235,7 @@ class CatAdapter:
             if reply_text and reply_text.strip():
                 self._consecutive_failures = 0
                 logger.info(f"🐱 Cat response for {cat_user_id}: {reply_text[:100]}...")
-                return reply_text
+                return reply_text, full_prompt
             else:
                 logger.warning("Cat returned empty response via WS")
                 self._consecutive_failures += 1