transcribe

Fetch a spoken-word transcript from a video URL. YouTube URLs try the uploader's captions first (seconds, no model download). Everything else — and YouTube videos without captions — falls through to audio extraction + mlx-whisper on Apple Silicon, with a headless-Playwright fallback for Threads and other yt-dlp-unsupported platforms.

When to invoke

• The user gives you a YouTube, Shorts, Instagram, TikTok, or Threads URL and asks for a transcript, summary, analysis, or quote
• Another skill needs the spoken content of a video to do its job (e.g. research-scout analyzing a reference video)

How to invoke

From the Bash tool:

transcribe <url>                  # default: captions-first for YouTube, whisper otherwise
transcribe <url> --json
transcribe <url> --force-whisper  # skip YouTube captions, always run whisper