Description
watch is a Claude Code skill that lets an agent watch a video from a URL or a local file path. It downloads the video with yt-dlp, extracts frames with ffmpeg at a scale chosen automatically for the length of the video, and gets a transcript. The transcript comes from the video's native captions when they exist, and otherwise from a chosen local or cloud transcription backend such as WhisperX. There are two engines. With the gemini engine, which needs a GEMINI_API_KEY, Google's video model watches the video and the report includes a timestamped answer for the agent to relay. With the local engine, the script produces timestamped frames and a transcript, and the agent views the frames and answers from that evidence. The skill is strict about one point: a transcript alone cannot establish visual facts, so anything about what appears on screen must come from the frames or the video model. The commands use python3 on macOS and Linux, and on Windows the skill asks you to check for Python 3.10 or later. It is installed with npx skills from the bradautomates/claude-video repository, and its security audits show Gen Agent Trust Hub and Socket passing, with Snyk showing a warning.…
Highlights
- Watches a video from a URL or local path
- Frames extracted with ffmpeg at an automatic scale
- Transcripts from captions or local or cloud backends
- Gemini engine or local engine
- Timestamped evidence for answers
Ratings & Reviews
No ratings yet
Be the first to rate watch.
People also visit
canvas-design
Create beautiful visual art in .png and .pdf documents using design philosophy
watch FAQ
What can the skill do with a video?
Download it from a URL or read a local file, extract timestamped frames, and get a transcript so the agent can answer questions about it.
What is the difference between the two engines?
The gemini engine needs a GEMINI_API_KEY and uses Google's video model to watch the video. The local engine produces frames and a transcript for the agent to review.
Are native captions used?
Yes. Native captions come first, and the local or cloud transcription backend is only a fallback.
Can a transcript answer questions about visuals?
No. The skill says transcript-only evidence cannot establish visual facts.
How do I install it?
Run npx skills add https://github.com/bradautomates/claude-video --skill watch.
What do the security audits show?
Gen Agent Trust Hub and Socket pass, and Snyk shows a warning.
What are the best watch alternatives?
Popular alternatives to watch include implement, brand-guidelines and canvas-design. See all watch alternatives