The Agent Skills Directory

[COMMAND_EXECUTION]: The execution rule in SKILL.md (python scripts/main.py --url {用户提供的视频链接}) interpolates user-provided input directly into a shell command template. If the video link provided by a user contains shell metacharacters (such as ;, &&, or |), an attacker could potentially execute arbitrary commands on the host system.
[EXTERNAL_DOWNLOADS]: The scripts/main.py script automatically downloads and installs external software packages (yt-dlp and ffmpeg) using package managers like pip, brew, apt, and choco. While these are legitimate tools, automated runtime installation of software from remote repositories introduces supply chain risks and unexpected system changes.
[COMMAND_EXECUTION]: The script scripts/main.py uses subprocess.run to execute system-level installation commands (apt install, brew install, choco install). Attempting to install system packages via apt or choco typically requires elevated privileges, which represents an intrusive behavior for an AI agent skill.
[DATA_EXPOSURE]: The skill is configured to save downloaded media files to the user's ~/Downloads/ directory. While this aligns with its stated purpose, the script creates and modifies files in a user-accessible area of the filesystem.

yt-dlp-downloader