The Agent Skills Directory

Data Exposure & Exfiltration (HIGH): The skill is designed to automatically scrape browser cookie databases to bypass YouTube's anti-bot protections.
Evidence found in README.md and TROUBLESHOOTING.md describes 'Smart Cookie Management' and 'Automatic Extraction' from Chrome, Firefox, Edge, and Safari.
This behavior allows the agent to access the user's active session tokens and potentially other sensitive data stored in the browser's profile directory.
External Downloads (MEDIUM): The install_as_skill.sh script downloads and installs several Python packages from PyPI (yt-dlp, pysrt, python-dotenv).
Since the source repository (lxmxhh/cc-skill-youtube-download) is not on the trusted list, these third-party dependencies are considered unverifiable.
Command Execution (MEDIUM): The installation script install_as_skill.sh executes potentially dangerous shell commands including recursive directory removal (rm -rf) and directory creation within the user's home path (~/.claude/skills/).
Indirect Prompt Injection (LOW): The skill processes untrusted input from YouTube, such as video URLs and potentially metadata (titles, descriptions).
Ingestion points: YouTube URL processing and metadata extraction (inferred from output format in README.md).
Boundary markers: None detected in the provided documentation or scripts.
Capability inventory: The skill uses yt-dlp which can execute subprocesses for merging video/audio files via FFmpeg.
Sanitization: No evidence of sanitizing external content before displaying it to the user or agent.

youtube-download