reproduce

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

• Third-party content exposure detected (medium risk: 0.30). Runtime workflow reads outsider-authored free text by fetching and parsing the target GitHub issue via gh issue view <issueNumber> --json title,body,labels,comments, where the issue body and comments are user-authored.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill clones and later pushes the target repository at runtime using "https://github.com/.git", which fetches remote code that will be built/run (npm install, vite build, deploy), so remote content can execute and control the agent's behavior.