ffmpeg-captions-subtitles

Pass

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation provides curl commands to download Whisper GGML model files (e.g., ggml-base.bin, ggml-medium.bin) from Hugging Face's official repository at huggingface.co. These downloads are from a well-known and established service for AI models and are necessary for the skill's transcription functionality.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external, potentially untrusted data through FFmpeg's subtitle and text overlay filters.
  • Ingestion points: The skill uses ffmpeg to read external subtitle files (.srt, .ass, .vtt), video metadata, and external text files via the drawtext=textfile= parameter. It also processes audio streams for live transcription.
  • Boundary markers: There are no explicit instructions or delimiters provided to the agent to treat the content of these external files as untrusted or to ignore any embedded instructions.
  • Capability inventory: The skill enables the agent to execute complex ffmpeg commands that involve reading from and writing to the local file system.
  • Sanitization: The instructions do not specify any sanitization, validation, or escaping of the content read from external files before it is processed by the filters or potentially reflected in the agent's output.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 7, 2026, 09:30 PM