It was only a matter of time before someone packaged design engineering expertise into a format AI coding agents could actually consume. On January 7, 2026, ibelick created ui-skills, an open-source project that does exactly that. By mid-July, the project had exploded to 4,881 GitHub stars and 205 forks, becoming one of the fastest-growing developer tools in the design engineering space. The pitch is deceptively simple: run npx ui-skills start in your terminal, and your AI coding agent gains instant access to 181 curated skills spanning typography, animation, accessibility, color systems, component architecture, and visual polish. No more generic UI output. No more prompting an AI to "make it look better" and hoping for the best.
The project, built in TypeScript and distributed as an npm package via Astro, is the brainchild of ibelick, a designer and developer whose GitHub profile shows a deep commitment to bridging the gap between design intent and code output. The rapid adoption reflects a genuine shift in how developers are using AI coding tools: instead of treating agents as black boxes that produce code on demand, developers are now injecting structured domain expertise into their agent's context before the first line of code is written. ui-skills is the most polished example of this pattern to date.
What ui-skills Actually Does
At its core, ui-skills is a CLI tool that routes AI coding agents through the right skill set for whatever task they are working on. The project comes with 181 individual skills organized into categories including baseline UI, motion design, design systems, accessibility, typography, color, animation, interaction design, and visual polish. Each skill is a structured SKILL.md file that contains specific rules, constraints, and patterns an AI agent can apply to its output.
For example, the baseline-ui skill enforces an opinionated UI baseline designed to prevent what the project calls "AI-generated interface slop." It requires agents to use Tailwind CSS defaults, use motion/react (formerly framer-motion) for JavaScript animations, rely on accessible component primitives like Base UI or Radix, never animate layout properties, and keep interaction feedback under 200 milliseconds. These are not vague suggestions. They are enforceable rules that an agent can check its output against.
The improve-ui skill takes a different approach. It audits an existing product surface against its own design evidence, identifies verified UI problems, and writes self-contained implementation plans for another agent to execute. The skill enforces a strict evidence standard: every finding requires a proven contract, runtime verification that the issue reaches the affected surface, and a specific correction. If any of the three proofs is missing, the finding is rejected. This level of rigor is rare in both human and AI-driven design reviews.
Other notable skills include fixing-accessibility, which audits and fixes HTML accessibility issues with a priority-ranked rule system from critical (accessible names, keyboard access) to medium (contrast, media); fixing-motion-performance, which enforces compositor-only animation properties; and design-first-ui-prompting, a skill contributed by Meng To that helps agents produce design-conscious output from the start. The full registry spans 181 entries covering everything from Three.js animation to SwiftUI patterns to WebGL landing pages.
How to Use It
Getting started with ui-skills requires nothing more than a terminal and a compatible AI coding agent. The project works with Claude Code, Cursor, Codex (OpenAI), GitHub Copilot, Windsurf, Gemini, Cline, and any other MCP-compatible coding agent.
The primary workflow is straightforward:
npx ui-skills start launches the CLI in interactive mode, which asks what you are working on and routes your agent to the appropriate skill files. Alternatively, you can browse specific categories: npx ui-skills categories lists all available skill categories. npx ui-skills list --category motion shows all skills in the motion category. npx ui-skills get baseline-ui retrieves a specific skill file for direct inspection.
Once installed, the skill files become part of your agent's project context. Every time you work on UI code, the agent can reference the relevant skills automatically, applying the same constraints and patterns that a senior design engineer would bring to the task. The project uses the SKILL.md convention that many AI coding agents natively understand, so there is no plugin or extension required.
Why It Matters for Solo Founders
The design engineer role is one of the fastest-growing in tech: the person who can both design and code well. For solo founders and small teams, hiring a dedicated design engineer is often out of reach. ui-skills packages that expertise into a format AI coding agents can consume directly, meaning founders who use AI tools to build their products can now get production-quality UIs without hiring a dedicated designer or spending weeks learning design fundamentals.
The CLI-as-skill-injection pattern is also significant beyond ui-skills itself. It represents a new paradigm where developers skill up their AI agents with domain-specific expertise rather than relying on generic prompting. Instead of asking an agent to "make this look good" (a prompt that produces wildly inconsistent results), a developer can install a skill that tells the agent exactly what good looks like in measurable terms: use these primitives, follow these spacing rules, keep animations under 200ms, respect reduced motion preferences, use accessible names on every interactive control.
For founders building with Cursor or Claude Code, ui-skills effectively adds a design engineer to the team for the cost of a single npm install. The skills are MIT-licensed, so they can be modified, extended, or combined to match a specific product's design language.
Comparison to Alternatives
ui-skills is not the only project attempting to shape AI agent output, but it is the most comprehensive. GitHub Copilot's custom instructions, Cursor's rules files, and Claude Code's CLAUDE.md all allow developers to inject project-specific guidance, but they operate at the project level and require manual authoring. ui-skills operates at the skill level, with pre-built, community-vetted patterns that cover the full range of design engineering. It also integrates across multiple agents simultaneously: a single npx ui-skills start works whether you are in Claude Code, Cursor, or Codex.
The project also maintains a web registry at ui-skills.com that serves as a searchable catalog of all available skills, topics, and compatible agents. The registry includes contributions from multiple community members, making it more than a solo project: it is becoming a shared library of design engineering knowledge that grows with each new skill submission.
Who This Is For
ui-skills is for anyone who uses AI coding agents to build user interfaces and cares about the quality of the output. That includes solo founders building MVPs who need production-quality UIs without a design budget, frontend developers who want a reliable design baseline they can trust, design engineers who want to codify their own expertise into reusable skills, and teams that want consistent UI output across multiple developers using different AI tools.
The project is MIT-licensed and free to use. It runs entirely in your terminal with no SaaS dependency and no data leaving your machine. The skill library is actively growing as more developers contribute their own design engineering knowledge. If you use AI to write frontend code and want the output to look like a human designer touched it, ui-skills is worth the two seconds it takes to run npx ui-skills start.
