At least five builders shipped the same Claude video skill

A post named a /watch-video skill for Claude Code on August 3, and five builders had already shipped it independently.

If you're building a video-ingestion skill for an agent, the design decision here is already settled. Redoing it from scratch means repeating work five people already finished.

The skill covers YouTube, Loom, Zoom, Vimeo and local files. It pulls a transcript, grabs frames at key moments, and runs a vision pass over them, per the post that named it.

Checking that claim against public code turned up at least five matching repos: alexlarcheveque/claude-watch, Newuxtreme/watch-video-skill, jordanrendric/claude-video-vision, bradautomates/claude-video, and taoufik123-collab/claude-watch.

Every one of them uses yt-dlp to pull the video and ffmpeg to slice it into frames. Then it hands the model timestamped frame-and-transcript pairs instead of raw footage. None of the five appear to have copied from each other.

What's not established: adoption numbers, or who's actually running these in production.