harvest

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md, the runtime workflow explicitly uses “this session's transcript only” (including user turns) from the harness transcript path, which contains outsider-authored free text, to recover the thread and decide what to harvest (i.e., it ingests attacker-submitted conversation content at runtime).