agent-watchdog

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

• Third-party content exposure detected (high risk: 0.75). The skill’s runtime workflow ingests “transcript path / thread URL / PR / comments / CI logs” via host tools to reconstruct the contract and audit evidence, which can include outsider-authored free text (e.g., PR/issue/discussion comments or transcript messages from other agents/users) into the LLM context.