transcribe-meeting

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

• Third-party content exposure detected (high risk: 0.80). This skill explicitly accepts and downloads arbitrary Google Drive URLs (see "Option B: Google Drive URL" in SKILL.md and scripts/download-gdrive.sh), then transcribes and uses that untrusted audio to generate summaries, action items, and to drive uploads/updates, so third-party content can materially influence agent actions.

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

• Potentially malicious external URL detected (high risk: 0.80). The vad.py script calls torch.hub.load("snakers4/silero-vad") which will fetch and execute code from the remote GitHub repo (https://github.com/snakers4/silero-vad) at runtime (and the pyannote Pipeline.from_pretrained pulls model code from Hugging Face at https://huggingface.co/pyannote/speaker-diarization-3.1), and the skill defaults to running silero VAD—so remote code/content is fetched and executed as a required dependency.