connect-agent

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md describes running LangWatch scenarios that send user-provided conversation messages to a registered HTTP agent endpoint (Step 6) and requiring trace adoption so the judge relies on spans rather than reply text (Step 4), meaning outsider-authored free text is ingested via the agent’s chat request body.