The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts scripts/frames.py and scripts/transcript.py utilize subprocess.run to invoke ffmpeg, ffprobe, and yt-dlp. These operations use argument lists to mitigate command injection, though they operate on user-supplied video sources.
[DATA_EXFILTRATION]: Audio data is sent to the ElevenLabs API (https://api.elevenlabs.io) for transcription. The skill manages the ELEVENLABS_API_KEY through environment variables or .env file discovery.
[PROMPT_INJECTION]: The skill processes external media content, creating a surface for indirect prompt injection. Ingestion points: Data is ingested through video frame extraction and audio transcription scripts. Boundary markers: The SKILL.md instructions guide the agent to verify causal hypotheses against transcript content. Capability inventory: The environment allows for subprocess command execution, file system writes, and network requests to external APIs. Sanitization: The skill lacks automated sanitization or validation for ingested transcript text.

video-understand