The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script in scripts/understand_video.py contains logic to automatically install the openai-whisper package via pip if it is not found on the system. This package is a well-known and trusted library from OpenAI for audio transcription.
[COMMAND_EXECUTION]: The skill executes ffmpeg and ffprobe via the subprocess module to extract frames and metadata from video files. It also attempts to run the whisper CLI tool if the Python library is unavailable.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted video and audio data.
Ingestion points: Untrusted video files are processed via the video positional argument in understand_video.py.
Boundary markers: The JSON output format documented in references/output-format.md does not utilize delimiters or specific instructions to prevent the model from obeying instructions embedded in the video's audio or visual content.
Capability inventory: The script performs file system writes (creating a frames directory) and executes external media processing commands (ffmpeg).
Sanitization: There is no evidence of sanitization or content filtering applied to the generated transcription text before it is passed to the agent.

video-understand