chat-with-anyone

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

• Third-party content exposure detected (high risk: 0.80). SKILL.md Workflow A (steps A2–A3) explicitly instructs the agent to search for and download YouTube videos and their SRT subtitles (using the youtube-downloader) and scripts/extract_ref_segment.py parse those public, user-generated subtitle and audio files to select reference speech that is then used to generate TTS, so untrusted third-party content from the open web is fetched, read, and directly influences downstream actions.