clip-hand-skill

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

• Third-party content exposure detected (high risk: 0.85). Runtime path: Phase 2/3 uses yt-dlp to fetch subtitles/metadata from the user-supplied URL (e.g., yt-dlp --write-auto-subs ... --skip-download ... -o "source" "URL" producing source.en.json3), and Phase 3/4 then parses that outsider-authored subtitle text/JSON into the agent’s LLM context for segment selection.