English · 简体中文 · 日本語 · Español · Français
ESLint for design. It finds the exact line, the measured wrong value, and the exact fix.
What it does · Install · Use it · What you get · Works in your editor · Good to know
What it does
Most design review tells you to improve your contrast. This tells you that .hero-subtitle on
line 14 of src/styles.css is #fbfbfb on #ffffff, which measures 1.03 to 1, and that
#767676 would pass. Same shape as a linter error, applied to design.
It reads your CSS, JSX, HTML and Tailwind classes. By default nothing is executed, nothing leaves your machine, and no key is ever needed. Findings come in two kinds. A located finding, which is most of what you act on, carries three things:
- Located. The file, the line and the selector.
- Measured. The real wrong value, not an adjective.
- Fixed. The exact
beforetoafterchange, plus the WCAG criterion or named UX law it rests on.
The rest are project-level, like "5 font families, use 1 or 2". They score and they name the problem, but there is no single line to point at.
Install
That is Claude Code. For any other agent that reads the Agent Skills format, install the 14 skills straight from this repo:
The command-line tool needs no install step: npx ui-ux-suite . runs it.
Or add it as a dev dependency
Requires Node 18 or newer. Zero runtime dependencies, so this pulls nothing else.
Use it
1. Point it at a project.
You get a ranked list of findings and a weighted score out of 10 across 12 dimensions. No config file, no setup step.
2. Read a finding. This is real output from the broken fixture shipped in this repo, which scores 3.7 out of 10 because it is supposed to:
Run it yourself with npx ui-ux-suite test/fixtures/planted-ux-problems from a checkout.
3. Put it in CI. Pick the output and the gate that fit your pipeline:
Every flag is in docs/cli.md.
What you get
- A ranked report in Markdown, JSON, HTML or SARIF, whichever the reader is a machine or a person.
- A score out of 10 across 12 weighted dimensions, with accessibility carrying the most weight.
- A CI gate, either an absolute bar with
--fail-underor a baseline that fails on a new finding or a score drop. - 14 skills and 16 tools your agent can call, so "audit this project's design" works in chat.
Works in your editor
| Agent | One-line install |
|---|---|
| Claude Code | claude plugin install ui-ux-suite@10x |
| Any of 70+ agents | npx skills add Aboudjem/ui-ux-suite |
| Codex, Gemini CLI, OpenCode, Pi | ./install.sh <agent> |
| VS Code and GitHub Copilot | ./install.sh copilot |
| Everything else | see docs/editors.md |
Works in Claude Code, Cursor, Codex, Copilot, Gemini CLI, and 70+ other agents through
npx skills add. The skills are Markdown, so they run on whatever model your editor points at.
Add it as an MCP server instead
Cursor, VS Code, Gemini CLI, Windsurf, Continue, OpenCode and Zed take the same command as a config entry, in JSON, TOML or YAML depending on the editor. The per-editor snippets, including the three that key it differently, are in docs/editors.md.
Good to know
[!IMPORTANT] It audits, it never edits. It never modifies a source file in the project you point it at, and a fix is printed as
beforetoafterfor you to apply. The only files it writes are report and baseline files, and only where you asked for one.
- Nothing leaves your machine. Pure Node built-ins, no API key, no telemetry. The static audit makes no network call, and deep mode only visits the URL you hand it.
- Static source is the default. Deep mode is opt-in, needs
playwright-coreplus@axe-core/playwright, and degrades to source findings when they are absent. - The numbers are gated.
npm testruns 356 tests, and the shipped broken fixture must keep producing findings that carry a file, a line and a fix.
Learn more
- docs/cli.md, every flag, exit code and recipe
- docs/editors.md, install and MCP config per editor
- docs/scoring.md, the 12 weights and how the score is built
- docs/science.md, the WCAG criteria and UX laws behind the findings
- docs/faq.md and docs/comparison.md
- CHANGELOG.md, CONTRIBUTING.md, LICENSE
MIT · Built by Adam Boudjemaa