motion-prompt: a clickable dictionary of motion names your agent can build from

The most common failure in agent-made motion is vocabulary. You can picture the opening you want but can't name it, so the prompt says 'make it more premium' and the agent guesses. Shohei Ohtaki (@showheyohtaki) started posting short videos that put names to motion-graphics and camera moves. His Oct 16 post said the series had passed a million impressions, and he open-sourced the whole thing (MIT, repo created Oct 14). It got picked up fast: his launch post passed 900 likes, @Voxyz_ai's English write-up on Oct 17 passed 900 likes, and @GitHub_Daily featured it today. It had 465 stars when we checked.

What's in it, per the README: six packs of 24 names each, 144 in total. They cover motion-graphics moves (two packs), camera moves and transitions, UI component names, landing-page and web effects, and photo looks, lighting and styles. Every entry has a number, a name, a one-line meaning, an example of how to ask for it, and a moving sample in the gallery. The names are useful even without installing anything. 'Slow dolly in toward her face; move the camera, do not zoom' is the README's own example, and it works as well in a video-model prompt as in a coding agent.

The skill is the clever part. Install it with npx skills add showheyohtaki/motion-prompt, tell Claude Code or Codex 'Open the motion-prompt library', and the gallery opens in your browser. Press Pick on the samples you want, add a short note such as 'for the opening title', and press Send to AI. The agent then fetches only those specs and their reference code and builds them into your project. If your agent doesn't have the skill, Copy picks puts the same info, with the demo and spec URLs, on your clipboard. Once you know the numbers you can skip the gallery: 'use #012 and #045 to make this product video look professional', or 'turn camera move #032 into an English prompt for Veo'.

There's also a make-your-own folder with copy-paste prompts and a render script for building the same kind of library for your own field or language. The name lists are only in English and Japanese for now. Our take: this is the missing layer between taste skills like claude-motion and the renderer. It should drop into HyperFrames or Remotion pipelines, but we haven't tested how cleanly its reference code ports to either. Spotted, not tested. Credit Shohei Ohtaki.