Video Captions
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided video and audio files using transcription engines. This represents an attack surface where the generated text output from external media could influence agent behavior. 1. Ingestion points: Video and audio files processed via whisper and related tools in SKILL.md and engines.md. 2. Boundary markers: No specific delimiters or warnings for the agent to ignore instructions embedded in transcripts are provided. 3. Capability inventory: The skill uses ffmpeg, whisper, and whisper_timestamped via CLI for processing. 4. Sanitization: No explicit transcript sanitization is mentioned.
- [EXTERNAL_DOWNLOADS]: Documents installation of common Python packages (openai-whisper, mlx-whisper, whisper-timestamped, stable-ts) from official registries.
- [DATA_EXFILTRATION]: Mentions optional cloud transcription services (AssemblyAI, Deepgram) which involve sending audio data to their APIs. This is documented as an opt-in feature requiring user-provided API keys.
Audit Metadata