fix: 20q answer dedup cache — prevent repeated answers

Add a rolling cache of the last 30 answers (persisted to
twentyq_cache.json) and pass the recent list to the model as an
explicit avoid clause. Also prompt the model to vary categories
each round. If the model still returns a cached answer it is
rejected and one retry is attempted automatically.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

matrixbot/commands.py

@@ -2730,13 +2730,39 @@ async def cmd_ac(client: AsyncClient, room_id: str, sender: str, args: str):
 # ===========================================================================
 
 _TWENTYQ_GAMES: dict[str, dict] = {}
+_TWENTYQ_RECENT_MAX = 30
+_TWENTYQ_CACHE_FILE = Path("twentyq_cache.json")
+
+
+def _load_20q_cache() -> list[str]:
+    try:
+        data = json.loads(_TWENTYQ_CACHE_FILE.read_text())
+        return data.get("things", [])
+    except Exception:
+        return []
+
+
+def _save_20q_cache(things: list[str]) -> None:
+    try:
+        _TWENTYQ_CACHE_FILE.write_text(json.dumps({"things": things}, indent=2))
+    except Exception as e:
+        logger.warning("Failed to save 20q cache: %s", e)
+
+
+_20q_recent: list[str] = _load_20q_cache()
 
 
 async def _generate_20q_thing() -> dict | None:
+    avoid_clause = (
+        f" Do NOT use any of these recently used answers: {', '.join(_20q_recent[-20:])}."
+        if _20q_recent else ""
+    )
     system_msg = (
         "You are generating a subject for a game of 20 questions. "
         "Pick a specific, well-known, concrete thing. Avoid overly obscure topics. "
         "Good categories: animal, famous person, place, everyday object, food, movie/show, fictional character. "
+        "Choose a DIFFERENT category each time — vary between animals, people, objects, places, food, etc."
+        + avoid_clause + " "
         "Respond with ONLY a JSON object — no markdown, no explanation. "
         '{"thing": "elephant", "category": "animal", "hint": "it\'s a living creature"}'
     )
@@ -2763,6 +2789,14 @@ async def _generate_20q_thing() -> dict | None:
                 category = parsed.get("category", "thing").strip()
                 hint = parsed.get("hint", f"it's a {category}").strip()
                 if thing and len(thing) > 1:
+                    # Check it's not a repeat (case-insensitive)
+                    if thing.lower() in [r.lower() for r in _20q_recent]:
+                        logger.warning("20q generated a cached answer '%s', regenerating", thing)
+                        return None
+                    _20q_recent.append(thing)
+                    if len(_20q_recent) > _TWENTYQ_RECENT_MAX:
+                        _20q_recent.pop(0)
+                    _save_20q_cache(_20q_recent)
                     return {"thing": thing, "category": category, "hint": hint}
     except Exception as e:
         logger.error("20q generation error: %s", e, exc_info=True)
@@ -2807,7 +2841,7 @@ async def cmd_20q(client: AsyncClient, room_id: str, sender: str, args: str):
         return
 
     await send_text(client, room_id, "🤔 I'm thinking of something...")
-    thing_data = await _generate_20q_thing()
+    thing_data = await _generate_20q_thing() or await _generate_20q_thing()
     if not thing_data:
         await send_text(client, room_id, "Failed to think of something. Try again!")
         return