Narrow the window for longer videos. If you only care about the first 2 minutes of a 30-minute talk, tell Claude:

/watch <https://youtube.com/[url]> --start 0:00 --end 2:00 Break down the opening hook and first impression.

This gives you denser frame coverage and costs fewer tokens. The skill auto-adjusts: short clips get nearly every frame, long videos get a sparser scan.

Bump resolution if there's text on screen. If the video has slides, dashboards, or code on screen that Claude needs to read:

/watch <https://youtube.com/[url]> --resolution 1024 What does the pricing slide show?

Default is 512px wide. 1024 is better for reading on-screen text.

It works with more than YouTube. The skill supports TikTok, Instagram, Vimeo, Loom, X (Twitter) video, and a few hundred other platforms. Same syntax — just paste the URL.

It also works with local files. If someone sends you a screen recording or a video file:

/watch ~/Downloads/competitor-demo.mp4 What's happening in this recording?