ffmpeg-skill

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill performs localized media processing by wrapping FFmpeg and FFprobe. It avoids shell injection vulnerabilities by passing command arguments as lists rather than strings and implements an explicit allowlist in scripts/batch.py (ALLOWED_STEP_SCRIPTS) to ensure only authorized skill tools are executed.
  • [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted external data in the form of subtitles and metadata. It proactively mitigates injection risks through sanitization functions like ass_escape in scripts/_ass_overlay.py and validate_color in scripts/_common/color.py, which prevent malicious text from being interpreted as executable filter logic or override commands.
  • [EXTERNAL_DOWNLOADS]: The skill supports optional local speech-to-text transcription. While libraries like faster-whisper may download model weights from Hugging Face on first use, this is a standard requirement for the tool's primary purpose and is clearly documented in SKILL.md and scripts/_common/asr.py.
  • [OBFUSCATION]: Analysis of the source code found no hidden payloads or deceptive encoding. The use of sentinel characters (e.g., U+E000) and zero-width spacers in scripts/_ass_overlay.py was found to be for legitimate formatting and placeholder purposes in ASS files.
  • [DYNAMIC_EXECUTION]: Dynamic library loading via ctypes is used appropriately for Windows API compatibility (path normalization), and dynamic imports are used to handle optional third-party Python dependencies without forcing them on the user.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 04:41 AM