mimic

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

• Third-party content exposure detected (medium risk: 0.30). The Ghost subagent instructions (references/ghost.md, lines 35–40) direct the model to execute GitHub API calls searching for PR comments and issues authored by the user (gh api -X GET search/issues), whose bodies and comment text contain outsider-authored free text from other contributors in those PR threads.