The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill's installation instructions include the use of sudo to install dependencies on Linux systems.
Evidence: sudo apt install ffmpeg in SKILL.md.
[METADATA_POISONING]: The skill contains metadata at the end of the documentation claiming to be maintained by 'Claude Code Skills Collection', which contradicts the provided author identifier 'jackspace'.
Evidence: Maintained by: Claude Code Skills Collection in SKILL.md.
[DATA_EXFILTRATION]: The skill provides commands to access sensitive browser data, specifically cookies, which could be used to bypass security controls or access session-restricted content.
Evidence: yt-dlp --cookies-from-browser chrome and yt-dlp --cookies-from-browser firefox in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process metadata and content from 1000+ untrusted external websites, which could contain malicious instructions intended to influence the agent's behavior during content extraction.
Ingestion points: External video URLs, titles, and metadata from platforms like YouTube, Twitter, and TikTok (referenced in SKILL.md).
Boundary markers: None identified; untrusted data is passed to command-line arguments without delimiters or warnings.
Capability inventory: Executes powerful shell commands via yt-dlp for downloading and post-processing (referenced in SKILL.md).
Sanitization: No evidence of input validation or output sanitization before processing external content.
[COMMAND_EXECUTION]: The skill is centered around executing powerful command-line tools (yt-dlp, ffmpeg, aria2c) that interact directly with the local file system and network.

youtube-downloader