convex-agents

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

• Third-party content exposure detected (high risk: 0.90). The workflow ingests outsider-authored free text via the sendMessage action’s args.message (user-provided chat input) into the agent’s LLM context as messages: [{ role: "user", content: args.message }], which is not authored by the operating user’s system prompts/files but by an external party at runtime.