The Agent Skills Directory

[SAFE]: The skill implements a read-only dependency preflight (scripts/check_dependencies.sh) that only suggests manual installation steps (via apt, mise, or pipx) to the user instead of executing them directly.
[SAFE]: Data handling is restricted to the local file system using mktemp for temporary storage of audio chunks and JSON metadata, which are cleaned up upon exit unless --keep-artifacts is specified.
[SAFE]: Network operations are limited to communicating with the official Groq API (api.groq.com) and resolving public URLs via yt-dlp.
[SAFE]: Sensitive environment variables like GROQ_API_KEY are used for authorization via curl but are never printed or logged to the console.
[SAFE]: Command execution involves standard media processing tools (ffmpeg, ffprobe) and URL resolution (yt-dlp) with arguments derived from user input, but it follows best practices like using -- to separate flags and quoting arguments.

groq-media-transcript