learn-from-video

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

• Third-party content exposure detected (medium risk: 0.30). In Phase 2 and Phase 5, the skill navigates to the user-supplied YouTube video page and scrapes the “Show transcript” timestamped transcript rows, ingesting free text authored on the video page (e.g., captions/transcript text) before generating checkpoints.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly navigates at runtime to the YouTube page URL "https://www.youtube.com/watch?v=<VIDEO_ID>&t=<segment.start_seconds>s" and scrapes the transcript to drive segment planning and checkpoint prompts, so external page content directly controls the agent's prompts.