Watch
Gives Claude eyes for video: yt-dlp download, ffmpeg frames, local transcript
Test report
- Verdict
- Tested · Works
- Score
- Tested
- Jul 17, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · a3b1d50
Ran it live on a 19-second YouTube clip: it produced 19 timestamped frames plus a caption transcript, and reading the frames showed the actual zoo scene matching the audio. Without the skill Claude has no video input at all, so the delta is capability-defining rather than incremental. The setup preflight worked as documented (silent exit 0 when ffmpeg/yt-dlp are present), and transcription is genuinely local — the only outbound traffic is yt-dlp fetching the video itself. Requires ffmpeg and yt-dlp on the machine; the bundled installer handles macOS via Homebrew and prints exact commands elsewhere. The docs are unusually honest, down to a security section that matches what the code actually does.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 5/5
- Triggers reliably 4/5
- Output vs. baseline 9/10
- Docs & honesty 5/5
What Watch does
Lets Claude actually watch a video from YouTube, TikTok, Vimeo, X, Instagram or a local file: it downloads with yt-dlp, extracts timestamped frames with ffmpeg, and pulls a transcript from native captions or a fully local Whisper fallback (no API key). Triggers whenever a user pastes a video URL or points at a local video file and asks what is in it.
How to install Watch
git clone https://github.com/mathiaschu/watch
mkdir -p ~/.claude/skills
cp -r watch ~/.claude/skills/watch
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Watch
-
/watchGives Claude eyes for video: yt-dlp download, ffmpeg frames, local transcript
It also activates on plain-language prompts like these:
-
Watch this YouTube video and tell me what the host is demonstrating -
What does she say at the 0:45 mark in this TikTok clip? -
Look at demo.mp4 on my desktop and summarize what's happening in it
Frequently asked questions
- Is the Watch skill free?
- Yes. The skill itself is free from mathiaschu/watch. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Watch work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 17, 2026. Verdict: Tested · Works. Ran it live on a 19-second YouTube clip: it produced 19 timestamped frames plus a caption transcript, and reading the frames showed the actual zoo scene matching the audio. Without the skill Claude has no video input at all, so the delta is capability-defining rather than incremental. The setup preflight worked as documented (silent exit 0 when ffmpeg/yt-dlp are present), and transcription is genuinely local — the only outbound traffic is yt-dlp fetching the video itself. Requires ffmpeg and yt-dlp on the machine; the bundled installer handles macOS via Homebrew and prints exact commands elsewhere. The docs are unusually honest, down to a security section that matches what the code actually does.
- What is the Watch SkillProof Score?
- 9.2/10 — installs cleanly 5/5, triggers reliably 4/5, output vs. baseline 9/10, docs & honesty 5/5.
- How do I install Watch?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Watch with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.