watch

Run `npx skills add mathiaschu/watch --all -g -y` in your terminal to install the skill globally, then run the bundled setup script once to install ffmpeg, yt-dlp, and a local Whisper engine.

Paste the video URL with your question, and the skill downloads it, extracts frames, pulls the transcript, and answers based on what is actually shown and said.

No. Transcription runs 100% on your machine using mlx-whisper on Apple Silicon or openai-whisper on CPU, so no audio ever leaves your device.

Yes. For login-gated sources like Instagram or X, it borrows cookies from a browser where you are already logged in, reading them live without storing or transmitting them.

Videos under 10 minutes give the best accuracy, since frame coverage is capped at 100 frames and 2 fps. For longer videos, focus on a specific section with start and end times.