The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes video transcripts which could contain adversarial instructions designed to influence the agent's behavior.
Ingestion points: Video transcripts read from transcript.json in SKILL.md (Step 3).
Boundary markers: None specific; the model is instructed to synthesize the content rather than use delimiters.
Capability inventory: Execution of ffmpeg, ffprobe, and yt-dlp via subprocess in scripts/frames.py and scripts/transcript.py.
Sanitization: No sanitization of transcript text is performed before it is processed by the model.
[COMMAND_EXECUTION]: The skill invokes external command-line utilities (ffmpeg, ffprobe, yt-dlp) using the subprocess module. These calls use argument lists rather than shell strings, which is a recommended practice to mitigate shell injection risks.

video-understand