Teaching agents taste: Emil Kowalski launches UI and motion skills
Emil Kowalski just open-sourced a collection of design and motion skills for AI agents. From fixing awkward easing curves to Apple-grade fluid transitions and mobile polish, the rules inject real design engineering taste into AI code generation.

emilkowalski/skills is an open-source library of design and motion rules built specifically for AI coding agents. Created by Emil Kowalski, a former design engineer at Vercel and Linear and the maker behind Sonner and Vaul, the project packages years of frontend polish into actionable skills that agents can immediately reference.
The core premise is simple: coding agents know how to write CSS and JavaScript, but they typically lack taste. They routinely reach for ease-in on an entrance animation when physics demand ease-out. They throw a harsh solid border onto a container instead of layering a soft semi-transparent shadow. None of these mistakes break the app, but together they produce interfaces that feel unmistakably cheap.
What the skills actually teach agents
The repository breaks down design engineering instincts into modular commands that any agentic workflow can consume.
At the center is emil-design-eng alongside animate, which force models to choose correct curves, timing, and properties right from the start. For mobile developers, animate-expo brings the same precision to React Native and Expo, keeping fluid transitions, bottom sheets, and haptics running strictly off the main JavaScript thread to eliminate dropped frames.
One of the most practical tools in the bundle is improve-animations. Instead of letting an agent randomly edit files, it audits your entire repository across eight distinct categories, including interruptibility, physics, frequency, and accessibility. It writes a prioritized list of self-contained plans into your codebase, allowing a cheaper model to run the actual implementations without letting a frontier model touch production files directly.
The collection also includes apple-design, which translates fluid motion and layout principles from Apple's WWDC sessions into web-ready patterns. Meanwhile, mobile-native fixes notorious web quirks on phones, such as sticky hover states, tap highlight flashes, the 100vh viewport bug, and auto-zooming input fields.
How to get started
The project is fully open source on GitHub. Developers can add it to their existing agent toolchain by running npx skills@latest add emilkowalski/skills in their terminal.
Because the package consists of structured rules, heuristics, and prompt definitions rather than heavy binaries, it carries zero hardware overhead. It works alongside any frontier language model that supports contextual skills or custom project instructions.
Why it matters
We think this is a blueprint for how domain expertise should be distributed in the AI era. Great taste is often assumed to be an intangible talent, but Kowalski proves that high-end interface feel can be broken down into concrete constraints and edge cases.
For solo developers and full-stack engineers who want polished consumer-grade apps without hiring an agency, this repository is an immediate upgrade. It acts as an automated design lead that catches sloppy animations and mobile styling traps before code hits staging.
Sources (1)
- github.com(opens in a new tab)Originalgithub.com