baloney-detection-kit

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md requires the agent to treat “whatever claim or argument the user is working on” and “any text the user supplies” as untrusted data it ingests and screens through gated phases, meaning outsider-authored free text from the user directly enters the runtime workflow.