bradautomates /
claude-video
A skill that turns a video URL or file into frames and a transcript for an agent
- 19.1k stars
- 1.1k stars since 3 Oct
- Python
- Updated 28 days ago
- Some experience
- Agent skills & plugins
Agents can't normally look at video. Watch fixes that by turning a URL or local file into evidence the agent can read: frames taken at points in time, plus a transcript. You ask questions and the agent answers with timestamps.
It has two engines. If you set a Gemini API key, the whole video, picture and sound, goes to Google's video model and its timestamped answer comes back. The local engine runs on your machine with ffmpeg and yt-dlp, extracts frames, and reads native captions first. For videos without captions it can fall back to local WhisperX, Groq or OpenAI transcription.
It installs as a plugin in Claude Code (desktop app, VS Code or terminal) or as a skill in Codex and other Agent Skills hosts. It does not work in Claude Chat or Cowork. It never switches engines on its own: if Gemini fails, it reports the error and you decide whether to rerun locally.
What it does
Frames plus transcript
The local engine extracts timestamped frames from a URL or file and pairs them with a transcript, so the agent can describe what is on screen.
Captions first
Native captions are used when they exist. Without them you can choose WhisperX (local, no key), Groq whisper-large-v3, OpenAI whisper-1, or none.
Optional Gemini engine
With a GEMINI_API_KEY, the full video goes to Google's model. YouTube links go straight to Google; other files are uploaded and then deleted. --start and --end limit the range.
Works in several agents
Install through Claude Desktop's Code sessions, VS Code, or the Claude Code terminal. Codex and other Agent Skills hosts are supported too.
Setup check
A bundled setup.py checks for Python, FFmpeg and yt-dlp, and can install missing media tools through Homebrew on macOS.
What you could build
- Summarize a screen recording or demo with timestamps for what appears on screen.
- Ask an agent what a YouTube tutorial covers and where each topic starts.
- Pull a transcript from a talk or lecture, using captions or speech-to-text.
- Review a short local video clip for visible details, such as a bug repro.
- Limit analysis to a time range of a long video with Gemini's start and end options.
Get started
In a Claude Code session, add the marketplace and install the plugin, then start a new session and run /watch:watch on a video. The first-run wizard asks about detail level and transcription fallback; a free Gemini key is the fastest route, or choose local.
/plugin marketplace add bradautomates/claude-video/plugin install watch@claude-videoCommands are copied from the project's README. Read it on GitHub for the full setup.
More in Agent skills & plugins
superpowers
A set of skills that makes coding agents plan, test and review before they write code
mattpocock/skills
Small, editable agent skills for planning, testing and reviewing code with AI
ECC
Agents, skills, hooks and memory that add an engineering workflow to coding agents
andrej-karpathy-skills
A CLAUDE.md file that tells Claude Code to ask first, keep code small, edit narrowly
anthropics/skills
Anthropic's example Agent Skills: folders of instructions Claude loads per task
ponytail
A skill that makes AI coding agents write the least code that works
gstack
Slash-command skills that give Claude Code roles like reviewer, QA and release engineer
addyosmani/agent-skills
A pack of 25 Markdown skills that guide AI coding agents from spec to ship
taste-skill
Agent skills that steer AI coding tools toward better layout, type and motion
archify
Agent skill that turns a description or a codebase into an interactive HTML diagram