video-transcript

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

• Third-party content exposure detected (high risk: 0.85). The required workflow downloads and processes a user-supplied remote video's subtitle/caption file (uploader subtitles via yt-dlp ... --write-subs or auto-captions via yt-dlp ... --write-auto-subs), and then vtt_to_paragraph.py reads that VTT as text (args.vtt_file.read_text(...)) and emits it into the agent’s LLM context via the cleaned paragraph output—an outsider-authored free-text source.