The Agent Skills Directory

[COMMAND_EXECUTION]: The script understand_video.py uses subprocess.run to execute external binaries including ffmpeg, ffprobe, and the whisper CLI tool for video decoding and transcription.
[EXTERNAL_DOWNLOADS]: The skill includes functionality to install the openai-whisper Python package via pip if it is not found on the local system. This process is interactive and requires user confirmation before proceeding.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted video files to generate transcripts and images that are subsequently analyzed by the agent, creating a surface for indirect prompt injection.
Ingestion points: User-provided video files processed via understand_video.py in the scripts directory.
Boundary markers: Results are returned in a structured JSON format as documented in references/output-format.md, though the transcript text itself is not sanitized for potential embedded instructions.
Capability inventory: The skill utilizes subprocess execution of media tools and has the ability to create directories and write files to the local disk.
Sanitization: No specific filtering or escaping is applied to the generated transcription or video metadata before it is provided to the agent.

video-understand