video-edit

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

• Third-party content exposure detected (high risk: 0.85). Outsider free text is ingested into the LLM via scripts/simple_video_edit.py → generate_metadata(): it builds a Claude prompt containing transcript_with_times derived from transcribe_video(video_path) (Whisper output from the input video’s speech, i.e., non-user-authored content) and sends it to client.messages.create(...).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill calls torch.hub.load(repo_or_dir='snakers4/silero-vad') (i.e. https://github.com/snakers4/silero-vad) at runtime in jump_cut_vad_singlepass.py to fetch and execute remote code for Silero VAD, which is a required runtime dependency that executes external code.