Skip to content
Get packs

/watch

Lets an AI agent like Claude watch a video by turning it into timestamped frames and a transcript.bradautomates/claude-video

GitHub Download v0.3.2
Stars
19.7k
Forks
2.1k
Licence
MIT
Language
Python
Last update
1 months ago
Type
Dev tool

What it is

Watch is a skill (plugin) for AI agents such as Claude Code and Codex. You give it a video URL or a local file, and it produces timestamped still frames and a transcript that the agent can read and answer questions about. It uses native captions when available, and can fall back to speech transcription or, optionally, a Google Gemini video model.

  • Ask the agent to summarize a short local video with timestamps
  • Pull the transcript of a public video that has native captions
  • Have the agent describe what is visible on screen at specific moments
  • Run speech transcription on a video without captions, if you set up a transcription option

Great for

  • Developers who use Claude Code or Codex and want the agent to review a screen recording or tutorial
  • People who want a timestamped summary of a video they can share with an agent
  • Users who can install Python and FFmpeg and are comfortable running setup steps
  • Anyone analyzing public YouTube videos that have captions

Not ideal for

  • People who want to use it in Claude Chat or Cowork, which the README says are not supported
  • Users who cannot install FFmpeg, Python 3.10+, or yt-dlp, since the skill does not bundle them
  • Users with private videos who want the Gemini path, because the README says local files are uploaded to Google
Step-by-step guide: how to use /watchInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    See the README

    You'll need: Python 3.10+ · FFmpeg/ffprobe · yt-dlp · Claude Code, Codex, or another Agent Skills host

    Install in Claude Code terminal (slash command inside session)

    /plugin marketplace add bradautomates/claude-video

    Install the watch plugin in Claude Code

    /plugin install watch@claude-video

    Install for other local agents via Skills CLI

    npx skills add bradautomates/claude-video -g --skill watch
  2. 2

    Run it

    • Install the skill using one of the methods in the README for your agent
    • Let the agent check dependencies with the bundled setup script and install any missing tools
    • Put a short video such as example.mp4 in a folder shared with the agent
    • Ask the agent to use the watch skill on that file and summarize it with timestamps
  3. 3

    Get the most out of it

    • Request transcript detail for public URLs with native captions, which avoids a download of the video
    • Add a GEMINI_API_KEY in ~/.config/watch/.env for the whole-video Gemini path
    • Use --engine local to force the local frame and transcript path
    • Set up speech transcription (local WhisperX on v0.3.0 or later, or Groq/OpenAI) for videos without captions

Downloads

Source code (.zip)main branch · straight from GitHubRelease v0.3.21 files incl. watch.skill

Strengths

  • Installs as an Agent Skills plugin in several hosts, including Claude Code and Codex
  • Uses native captions first, so no transcription key is needed for many videos
  • Offers a no-key local path that extracts frames and transcripts on your machine

Trade-offs

  • Requires several external tools (Python 3.10+, FFmpeg, yt-dlp) and possibly a JavaScript runtime for YouTube
  • The Gemini path uploads local files to Google, which matters for private content
  • Does not work in Claude Chat, Cowork, or browser apps according to the README

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.