The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The installation instructions for Ubuntu/Debian use sudo apt-get install. Requiring or encouraging the use of sudo within a skill is a privilege escalation risk, as it grants the execution environment administrative access to the system.
[EXTERNAL_DOWNLOADS] (MEDIUM): The skill installs rmbg-cli globally using npm install -g. This package is not from a trusted organization defined in the security scope, introducing a supply chain risk through unverified third-party code.
[COMMAND_EXECUTION] (MEDIUM): The skill executes shell commands (ffmpeg, magick, mogrify, rmbg) using direct filename interpolation. While the examples show static filenames like input.png, in practice, an agent might pass user-supplied filenames. Without explicit sanitization or shell-escaping, this pattern is vulnerable to command injection if a filename contains shell metacharacters (e.g., file; rm -rf / .png).
[COMMAND_EXECUTION] (LOW): The skill processes untrusted media files using tools with complex parsers (FFmpeg, ImageMagick). Historical vulnerabilities (e.g., ImageTragick) demonstrate that malicious media files can lead to remote code execution. No sanitization or sandboxing guidance is provided for these ingestion points.

media-processing