The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/understand_video.py uses the subprocess module to execute ffmpeg and ffprobe for frame extraction and video metadata probing. Commands are executed using argument lists, which prevents shell injection.
[EXTERNAL_DOWNLOADS]: The skill offers to install the openai-whisper package via pip if it is not found on the system. This installation is performed within the _check_and_offer_install function and requires user confirmation in an interactive session.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (video files) to extract transcripts and frames.
Ingestion points: Video files provided by the user are processed by ffmpeg and Whisper in scripts/understand_video.py.
Boundary markers: The output is structured as a JSON object, providing some separation between metadata and transcribed text.
Capability inventory: The skill can execute shell commands (ffmpeg) and write results to local files.
Sanitization: The transcribed text is extracted directly from the audio and included in the output without sanitization, which could allow a video with malicious audio/subtitles to influence the agent's behavior.

video-understand