The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the yt-dlp binary using the subprocess.run method to extract video metadata and download media files.
Evidence: Implementation found in scripts/download.py within get_formats and download_video functions.
Security context: The implementation uses the list-based argument passing method which prevents shell-level command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill downloads video and audio content from external hosting platforms (YouTube, Instagram, etc.) as its primary function.
Evidence: Calls to yt-dlp with user-supplied URLs in scripts/download.py.
[PROMPT_INJECTION]: The skill processes untrusted metadata (such as video titles and notes) returned by external platforms, presenting an indirect prompt injection surface.
Ingestion points: scripts/download.py (via yt-dlp metadata extraction displayed to the user).
Boundary markers: Absent in the data extraction logic.
Capability inventory: subprocess.run for media processing and CLI-based file removal (rm) instructed in the skill workflow.
Sanitization: Output filenames are sanitized via whitelist regex (re.sub(r"[^a-zA-Z0-9._-]", "_", ...)), though descriptive text displayed to the agent is not explicitly filtered for embedded instructions.

universal-video-downloader