ai-agent

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The ai-agent runtime loop consumes outsider-authored user text from the actor’s message queue (connection.send("message", { text, sender }) → for await (const queued of c.queue.iter()) → model prompt), so arbitrary connected clients can inject free text into the LLM input.