The Agent Skills Directory

Remote Code Execution (CRITICAL): The skill contains explicit instructions to install Homebrew by fetching a script from the internet and piping it directly into bash (/bin/bash -c "$(curl -fsSL ...)"). This allows for unauthenticated arbitrary code execution from a remote, non-trusted source.
Unverifiable Dependencies (HIGH): The instructions require the agent to automatically install yt-dlp and ffmpeg via brew install without seeking user confirmation. This bypasses typical security boundaries and introduces software of unknown integrity into the host system.
Indirect Prompt Injection (HIGH):
Ingestion points: The skill processes arbitrary URLs provided by users which point to external web content (YouTube, Twitter, Bilibili, etc.) in SKILL.md.
Boundary markers: No delimiters or boundary markers are used when interpolating user-provided URLs into shell commands.
Capability inventory: The skill uses yt-dlp which executes shell processes, interacts with the local file system, and performs network requests.
Sanitization: There is no evidence of input validation or sanitization to prevent shell metacharacter injection or exploitation of yt-dlp metadata parsing vulnerabilities.
Command Execution (MEDIUM): The skill includes automated logic to modify the system state (installing packages, checking versions) every time it is used, which increases the attack surface for local privilege escalation.

video-downloader