The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes external CLI tools (yt-dlp, ffmpeg, and ffprobe) to process video data.
Evidence: The script capture_video_frames.py uses list-based arguments in subprocess.run(), avoiding shell injection vulnerabilities.
Evidence: User-supplied youtube_url arguments are validated against a specific regular expression in the extract_video_id function before being used in commands.
[PROMPT_INJECTION]: The skill uses a subagent to generate descriptions which are then reused in subsequent prompts, creating an indirect prompt injection surface.
Ingestion points: The agent reads frames_manifest.md, which contains previously generated descriptions, to provide context for the next frame's analysis.
Boundary markers: The prompt template in SKILL.md wraps the potentially untrusted previous description in double quotes: "[PREVIOUS_DESCRIPTION]".
Capability inventory: The skill can execute file writes and run local CLI tools via the provided Python script.
Sanitization: No specific escaping or content filtering is applied to the frame descriptions before they are interpolated into new prompts.

capture-video-frames