audio-transcribe

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

• Third-party content exposure detected (high risk: 0.80). The SKILL.md explicitly allows --in to be an audio file URL (e.g., "Input audio file path or URL" and examples like --in interview.mp3), so the agent can fetch and transcribe arbitrary public/untrusted audio which it will read/interpret as part of its workflow, enabling indirect prompt injection.