more optmizations

web/agent/src/agent.py

@@ -23,9 +23,9 @@ logger.info("Initializing Cavepedia agent...")
 
 
 def limit_history(ctx: RunContext[None], messages: list[ModelMessage]) -> list[ModelMessage]:
-    """Limit conversation history to manage token usage and request size."""
-    # Keep only the last few messages to avoid large requests hitting Cloudflare limits
-    return messages[-4:]
+    """Limit conversation history to manage token usage."""
+    # Keep last 8 messages for context, but not unlimited
+    return messages[-8:]
 
 def check_mcp_available(url: str, timeout: float = 5.0) -> bool:
     """Check if MCP server is reachable via health endpoint."""
@@ -54,7 +54,7 @@ Rules:
 4. Can create ascii diagrams/maps.
 5. Be direct—no sycophantic phrases.
 6. Keep responses concise.
-7. Use tools sparingly—one search usually suffices. Answer from your knowledge when possible."""
+7. Search ONCE, then answer with what you found. Do not search repeatedly for the same topic."""
 
 
 def create_agent(user_roles: list[str] | None = None):