2 boosters for "llm-inference" — open source, verified from GitHub, ready to install
Heuristic scoring (no AI key configured).
You are an output safety guard for Captain Claw. "verdict": "allow" | "suspicious", "reason": "short reason"
