/watch
A Claude Code skill that downloads a video from YouTube, Instagram, X, Vimeo or almost any yt-dlp-supported site, or a local file, extracts frames and transcribes the audio locally, so Claude can read the frames as images and answer questions about the video.
/watch is a Claude Code skill that gives Claude the ability to watch a video. Given a URL from YouTube, Instagram, X/Twitter, Vimeo, TikTok, Loom or roughly 1,800 other yt-dlp-supported sites, or a local video file, it downloads the video, extracts frames, and produces a timestamped transcript, then has Claude read each frame as an image alongside the transcript.
Transcription runs entirely on the local machine, using mlx-whisper on Apple Silicon or openai-whisper on CPU, so no API key or paid transcription service is required and no audio leaves the machine. Browser cookie support lets it fetch from login-gated sources such as Instagram and X.
It installs as a Claude Code plugin or as a skill folder for other agent harnesses such as Codex, and requires Python 3.9 or newer plus ffmpeg and yt-dlp, which a setup script installs automatically.