AI agents now read mail, consume untrusted content, and send messages for us — Simon Willison's *lethal trifecta*: one poisoned message can steal private data. Trifecta Checker is a chat with Trafica, a friendly checker. Describe your agent or drop in an MCP config; Muse Spark builds a capability profile, a deterministic rule engine (never the LLM) rates it Low / Risky / Lethal, and Trafica explains it plainly, with gentle fixes. Six languages. Built for Meta's Global AI Developer Hackathon.