claude-watch

Run `npx skills add devinilabs/claude-watch --all -g -y` in your terminal to install it globally. It requires ffmpeg and yt-dlp, which the built-in setup script can install automatically on macOS.

Paste the video URL with an optional topic, and claude-watch downloads it, extracts scene-based frames, pulls the transcript, and writes a structured notes file with screenshots, key concepts, and code blocks.

Yes. Videos with native captions are processed completely free using yt-dlp and ffmpeg. A Groq or OpenAI key is only needed as a Whisper fallback when a video has no captions.

Yes, but best accuracy is under 30 minutes per pass. For longer videos, use the --start and --end flags to focus on the section you care about.

Notes are saved to a persistent library at ~/claude-watch/library/<slug>/ with the video, transcript, and frames cached. Re-running the same URL reuses the cache and only regenerates notes.