Member-only story
Paste a YouTube link into an AI and it basically guesses from the title. A free open-source skill called /watch changes that — by slicing video into the two things AI can actually see. #
Here’s a gap in AI tools you’ve probably bumped into without naming it. Claude can read a webpage, write and run code, browse a whole repository — but paste it a YouTube link, and it’s suddenly useless. As the creator of this tool puts it in his own README: out of the box, Claude “has to either guess from the title or pull a transcript that’s missing 90% of what’s on screen.”
Think about how much information lives in video that a transcript never captures: the code on the presenter’s screen, the exact moment a UI breaks in a bug recording, what’s actually shown during the first three seconds of a viral clip. All invisible to your AI.
A developer named Brad Bonanno — a YouTuber who builds AI automation for businesses — got tired of this and shipped a fix: an open-source skill called
(repo: bradautomates/claude-video, MIT license). It's free, it's picked up around 2,500 GitHub stars, and the way it works is a genuinely clever…/watch