Fix memory system: lower thresholds, add dedup, store from free-mode

.claude/skills/mobius-judge/scripts/record_verdict.py

@@ -15,7 +15,10 @@
 sys.path.insert(0, "src")
 
 from mobius.config import get_config
-from mobius.db import init_db, row_to_dict
+from mobius.db import init_db, row_to_dict, vec_to_blob
+from mobius.embedder import embed
+from mobius.memory import Memory
+from mobius.models import MemoryEntry
 from mobius.registry import Registry
 from mobius.tournament import Tournament
 
@@ -39,9 +42,10 @@ def main():
     reasoning = args[2]
 
     config = get_config()
-    conn, _ = init_db(config)
+    conn, vec_available = init_db(config)
     registry = Registry(conn, config)
     tournament = Tournament(conn, config, registry)
+    memory = Memory(conn, config, vec_available)
 
     # Get the match
     if match_id:
@@ -122,6 +126,21 @@ def main():
 
     conn.commit()
 
+    # Store in vector memory so future selections benefit
+    task_text = match.get("task_description", "")
+    if task_text and full_winner_id:
+        try:
+            task_vec = embed(task_text, config)
+            memory_entry = MemoryEntry(
+                task_embedding=vec_to_blob(task_vec),
+                task_text=task_text,
+                winning_agent_id=full_winner_id,
+                score=max(scores.values()) if scores else 0.0,
+            )
+            memory.store(memory_entry)
+        except Exception as e:
+            print(f"Warning: failed to store memory entry: {e}", file=sys.stderr)
+
     # Print results
     winner = agents_by_id.get(full_winner_id)
     print(f"Verdict recorded for match {mid[:8]}")

src/mobius/config.py

@@ -23,6 +23,7 @@ class MobiusConfig(BaseModel):
     agent_timeout_seconds: int = 120
     agent_max_turns: int = 10
     agent_budget_usd: float = 0.05
+    agent_max_output_tokens: int = 16384
 
     # Judge
     judge_models: list[dict[str, str]] = [
@@ -41,8 +42,8 @@ class MobiusConfig(BaseModel):
     embedding_model: str = "all-MiniLM-L6-v2"
     embedding_dim: int = 384
     memory_top_k: int = 5
-    similarity_specialist_threshold: float = 0.9
-    similarity_ensemble_threshold: float = 0.7
+    similarity_specialist_threshold: float = 0.5
+    similarity_ensemble_threshold: float = 0.3
 
     # Self-improvement
     max_agent_population: int = 50

src/mobius/memory.py

@@ -35,7 +35,18 @@ def __init__(
         self.vec_available = vec_available
 
     def store(self, entry: MemoryEntry) -> None:
-        """Store a task outcome in memory."""
+        """Store a task outcome in memory, skipping duplicates."""
+        existing = self.conn.execute(
+            "SELECT id FROM memory WHERE task_text = ? AND winning_agent_id = ?",
+            (entry.task_text, entry.winning_agent_id),
+        ).fetchone()
+        if existing:
+            logger.debug(
+                "Duplicate memory entry for agent %s on task, skipping",
+                entry.winning_agent_id,
+            )
+            return
+
         row = dict_to_row(entry.model_dump(exclude={"task_embedding"}))
         cols = ", ".join(row.keys())
         placeholders = ", ".join(["?"] * len(row))

src/mobius/providers/anthropic.py

@@ -75,7 +75,7 @@ async def _run_simple(
         try:
             response = await asyncio.wait_for(
                 client.messages.create(
-                    model=model, max_tokens=4096,
+                    model=model, max_tokens=16384,
                     system=system_prompt,
                     messages=[{"role": "user", "content": prompt}],
                 ),
@@ -115,7 +115,7 @@ async def _run_with_tools(
             for turn in range(max_turns):
                 response = await asyncio.wait_for(
                     client.messages.create(
-                        model=model, max_tokens=4096,
+                        model=model, max_tokens=16384,
                         system=system_prompt,
                         messages=messages,
                         tools=[ANTHROPIC_BASH_TOOL],

src/mobius/providers/google.py

@@ -123,7 +123,7 @@ async def _run_with_tools(
         config = types.GenerateContentConfig(
             system_instruction=system_prompt,
             tools=[bash_tool],
-            max_output_tokens=4096,
+            max_output_tokens=16384,
         )
 
         contents = [types.Content(

src/mobius/providers/openai.py

@@ -65,7 +65,7 @@ async def _run_simple(
                         {"role": "system", "content": system_prompt},
                         {"role": "user", "content": prompt},
                     ],
-                    max_tokens=4096,
+                    max_tokens=16384,
                 ),
                 timeout=timeout_seconds,
             )
@@ -110,7 +110,7 @@ async def _run_with_tools(
                         model=model,
                         messages=messages,
                         tools=[OPENAI_BASH_TOOL],
-                        max_tokens=4096,
+                        max_tokens=16384,
                     ),
                     timeout=timeout_seconds,
                 )

src/mobius/providers/openrouter.py

@@ -78,7 +78,7 @@ async def _run_simple(
                         {"role": "system", "content": system_prompt},
                         {"role": "user", "content": prompt},
                     ],
-                    max_tokens=4096,
+                    max_tokens=16384,
                 ),
                 timeout=timeout_seconds,
             )
@@ -123,7 +123,7 @@ async def _run_with_tools(
                         model=model,
                         messages=messages,
                         tools=[OPENAI_BASH_TOOL],
-                        max_tokens=4096,
+                        max_tokens=16384,
                     ),
                     timeout=timeout_seconds,
                 )