diagnose

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

• Third-party content exposure detected (medium risk: 0.65). The required workflow is a debugging playbook that instructs the agent to run user-provided loops (e.g., “HITL bash script” and “captured trace”); in the HITL path, the script explicitly captures free-form user responses (ERROR_MSG) and prints them for the agent to parse, which is outsider-authored text entering the LLM context.