Repolode
Submit
bradautomates

bradautomates/claude-video

A Claude Code and Agent Skills plugin that lets AI agents watch videos by turning a URL or local file into timestamped frames and a transcript.

View on GitHub
Stars
18,826
Forks
1,959
Open issues
208
Last commit
25 days ago

What is it?

Watch is an agent skill that gives tools like Claude Code, Codex, and other Agent Skills hosts the ability to review video. You pass it a URL or local file, and it produces timestamped frames and a transcript that the agent can read and summarize. Native captions are used first when available.

It offers two engines. The Gemini engine sends the whole video to Google's video model and relays a timestamped answer, with YouTube links needing no extra tools. The local engine runs on your machine using ffmpeg and yt-dlp to extract frames and captions, and can add speech transcription through local WhisperX or cloud Groq and OpenAI Whisper APIs when a video has no captions.

The skill controls frame sampling by detail level, time range, and frame cap, and it removes near-duplicate frames. It also supports cookies for authenticated sources and includes a setup script that checks required tools such as Python 3.10+, FFmpeg, and yt-dlp.

Key features

  • 01Turns a URL or local video into timestamped frames and a transcript.
  • 02Uses native captions first and falls back to speech transcription when none exist.
  • 03Offers a Gemini engine that analyzes the full video, plus a fully local engine.
  • 04Supports local WhisperX or Groq/OpenAI Whisper for transcription.
  • 05Controls frame count, detail level, and time range, with duplicate-frame removal.
  • 06Installs in Claude Code, Codex, and other Agent Skills hosts.

Use cases

  • Having an AI agent summarize a tutorial or talk with timestamped references.
  • Asking an agent to inspect a specific segment of a screen recording or bug demo.
  • Extracting a transcript and key on-screen visuals from a YouTube video for analysis.
  • Reviewing local meeting or product recordings without manually scrubbing through them.

Quick start

  1. 1

    Add the marketplace in Claude Code

    /plugin marketplace add bradautomates/claude-video
  2. 2

    Install the Watch plugin

    /plugin install watch@claude-video
  3. 3

    Start a new Claude Code session and run the skill on a video

Things to know

  • Watch does not work in Claude Chat or Cowork, only in Claude Code, Codex, and other Agent Skills hosts.
  • Without a Gemini key it needs local ffmpeg and yt-dlp, and YouTube also needs a JavaScript runtime.
  • Speech transcription needs a WhisperX setup, or a Groq or OpenAI API key.
  • Local WhisperX requires Watch v0.3.0 or later.

Topics

#claude-code#agent-skills#video-analysis#transcription#ffmpeg#yt-dlp#whisper#multimodal

FAQ

What is claude-video?
A Claude Code and Agent Skills plugin that lets AI agents watch videos by turning a URL or local file into timestamped frames and a transcript. Watch is an agent skill that gives tools like Claude Code, Codex, and other Agent Skills hosts the ability to review video. You pass it a URL or local file, and it produces timestamped frames and a transcript that the agent can read and summarize. Native captions are used first when available.
Is claude-video free to use? What license does it have?
claude-video is open source under the MIT license. Check the license terms before using it in commercial products.
Is claude-video still maintained?
Actively maintained. The last commit to claude-video was on Sep 25, 2026.
What can I use claude-video for?
Having an AI agent summarize a tutorial or talk with timestamped references. Asking an agent to inspect a specific segment of a screen recording or bug demo. Extracting a transcript and key on-screen visuals from a YouTube video for analysis. Reviewing local meeting or product recordings without manually scrubbing through them.
How do I install claude-video?
Add the marketplace in Claude Code: /plugin marketplace add bradautomates/claude-video Install the Watch plugin: /plugin install watch@claude-video Start a new Claude Code session and run the skill on a video
Where can I get claude-video?
The source code and installation instructions are on GitHub: https://github.com/bradautomates/claude-video

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects