The 6 skills Meng To commonly uses—his demos rely on them for visual polish

Meng To listed 6 skills he has been using on X, for making demos. klöss reposted it. The list isn't long, and the problem it points to is very focused: a demo's functionality can be faked, but its visual feel can't.

First, some context. A skill is a package of instructions and resources that an agent can load. Once installed, you type a slash command in the prompt, and it works through a preset workflow. The difference from an ordinary prompt is that it can carry files, scripts, and a set of evaluation criteria.

Meng To's demo video from the original post:

dream-loop

By anshuc. The approach is a closed loop:

1. First, AI uses image generation to 'dream' a high-quality target screenshot.
2. Build toward that target image.
3. Another critic agent compares the live screenshot with the target image and gives feedback.
4. Go back to step 2 and loop until the critic is satisfied.
5. Optionally, go back to step 1 and dream a better target based on the current state.

If you want 3D, install Blender; using MCP or a script interface works better than having the agent operate the UI. The author says it has only been tested so far in Codex with GPT-6 Astra, but strong models like Claude Fable 5.1 can very likely run it too.

Install: npx skills add achimala/dream-loop

Online demo: https://dream-loop-demo.anshu.dev

Hairline

Made by Lucas Marques. 33 isometric line-art graphics, pure SVG, no dependencies, works with both React and native DOM. The selling point is in the page title: the line art 'responds' to the pointer. Wherever the mouse moves, the graphic reacts.

npm i @lucasmarkes/hairline, MIT license. It also provides a set of skills so you can draw according to your own needs.

Emil Kowalski's skills

The link is on emilkowal.ski/skill and actually resolves to aiforui.dev. 18 skills, ordered by what you'd actually encounter in a project: motion vocabulary, choosing a UI library, prototyping, design engineering, animation, choreography, Apple-style design, gestures, mobile native, animation performance, stress-testing the UI with bad data, auditing animation, finding motion opportunities, improving, removing unnecessary animation.

56k stars on GitHub, 3.4 million installs on skills.sh. Install command: npx skills@latest add emilkowalski/skills.

What's worth looking at on this page is the example phrases. It pairs things users would say with the corresponding skill, e.g. 'Our toolbar tooltip feels sluggish—can you fix it?' maps to /emil-design-eng; 'When I delete a card, the other cards jump—fix it' maps to /choreograph; 'What spring should this drawer use?' maps to /apple-design.

Meng To's own skills

github.com/mengto/skills, a set for designers and builders. The content isn't expanded on; from the naming, it looks like a workflow to support his usual demo-making.

REA

rea.tools. Reverse-engineer any app or website, take it apart to see how it's implemented. When searching for references for a demo, it's more reliable than screenshotting and guessing.

HTML Demo

fil.design/skills/html-demo. Get your app running directly in HTML. Seeing the result is faster than reading a description.

A few observations

Of the 6, at least 4 solve a 'taste' problem, not a 'capability' problem. Models have long been good enough at writing code; where they get stuck is not knowing what the right feel is, or what it means for an animation to be one notch too much.

So the distribution format of skills is changing too. These things used to be scattered across blogs, tutorials, and design-system docs; now it's one line of npx to install into an agent, and a slash command in the prompt. Emil's 3.4 million installs show one thing: what people lack isn't framer-motion API docs, it's 'what value should this spring actually be?'