bradautomates/claude-video

Project overview

This project adds video comprehension to AI coding assistants and chat agents by downloading content, extracting frames, and transcribing audio. It operates as a portable skill unit across over 50 agent hosts without modification, giving models grounded context from visual and textual video data.

Project type

AI Agent · Video

Use cases

Data Analysis

Deployment

Refer to project documentation

License

MIT

Best for

• Developers and general users needing to query video content through API-driven AI assistants like Claude Code, Cursor, or claude.ai.

Key capabilities

• Uses a yt-dlp wrapper to check native captions first and download only what is needed for the run, supporting YouTube, Loom, TikTok, X, Instagram, and local paths.
• Extracts image frames from videos using ffmpeg with auto-fps logic and a deduplication pass to drop near-identical frames.
• Parses VTT native captions first, falling back to Whisper orchestration via Groq or OpenAI APIs.
• Checks for dependencies (ffmpeg, yt-dlp) on the first run and automatically installs them or provides exact install commands.

Limitations and risks

• Long-video accuracy depends on the detail mode; capped modes thin out past approximately 10 minutes.

Getting started

• Install the skill via plugin marketplace or npx skills add. Dependencies are auto-installed on the first run. Use the /watch command with a URL and a question in Claude Code to begin.

Evidence and sources

• GitHub project description: Give Claude the ability to watch any video. /watch downloads, extracts frames, transcribes, hands it all to Claude.
• README: The CLI discovers the skill from `skills/watch/SKILL.md` and copies the whole folder — `SKILL.md` plus its `scripts/` runtime — as a self-contained unit.
• README: Claude can read a webpage, run a script, browse a repo. What it can't do, out of the box, is *watch a video*.
• README: you can paste a URL or a local path, ask a question, and Claude fetches captions first, downloads only what it needs, extracts frames (scene-aware, or fast keyframes at `efficient…
• README: URL (anything yt-dlp supports — YouTube, Loom, TikTok, X, Instagram, plus a few hundred more) or a local path

Reviewed at: 2026-07-18T04:23:21Z