ui-ux-pro-max-skill-cn: a translation, and what it tells you about the tool

Seven files, and the subject is upstream

Start by being clear about what this repository is, because the name makes it look like a fork and it is not.

The complete file list is a CNAME file for a GitHub Pages custom domain, a licence, two READMEs where one is described as the older Chinese version, and three web files: a page, a stylesheet and a script. That is the whole repository. There is no code here implementing anything about design.

The subject is a project that lives elsewhere, and this site is its Chinese tutorial, translating and localising the upstream README. The README is explicit that the site has been synced with the upstream v2.0 update, and the badge row carries the version it tracks. So the value of this repository is documentation in a language the upstream does not publish, and the value of the upstream is the thing you would actually install.

That framing matters for two reasons. If you are deciding whether to use the skill, this site is a good place to read a long, specific description of what it does, because a translation tends to preserve detail that a summary would drop, including the exact category names and the exact counts. And if you are deciding whether to trust the documentation, remember that it is maintained by hand: a stat row, a command list and an install method all have to be copied across when upstream changes, and there is no check that they match.

The licence is MIT and the site is served from a custom domain recorded in the CNAME file. The last push was on 2026-06-29.

One detail worth noting for anyone reading the numbers: the file list contains a second README described as the older Chinese description, which is the sort of leftover that accumulates when a document is restructured and the old one is kept rather than removed.

161 product categories, and a BM25 ranker behind them

The taxonomy is the interesting part of this tool, and the tutorial reproduces it in full.

The reasoning rules are organised by industry, and the categories run from Tech and SaaS, through Finance, Healthcare, E-commerce and Services, into Creative, Lifestyle and something called Emerging Tech. Within those, the product types are specific enough to be useful: a medical clinic and a veterinary practice are separate entries, as are a personal finance tracker and an invoice and billing tool, and in the emerging category you find Web3, spatial computing, quantum computing and a category for an autonomous drone fleet.

That last one is the detail that tells you the taxonomy was built by covering ground rather than by designing a model. It is also why the count matters: the description gives 161 colour schemes aligned one-to-one with 161 product types, so asking for a design for a drone fleet is a lookup rather than an inference, and that alignment is the mechanism by which the tool avoids generic output.

Now the retrieval, because it explains how the output is assembled. The generation flow has four steps. A user request arrives. Then a multi-domain retrieval runs, matching in parallel across product type, UI style, colour, landing page patterns and font pairings. Then a reasoning engine filters the candidates using product classification rules, a BM25 ranking, industry anti-patterns and conditional rules. Then a full output is generated covering pattern, style, colours, typography, effects, anti-patterns and checklist.

BM25 is a lexical ranking function, which means the retrieval matches words rather than meaning. That has a direct consequence for the design: step two is five parallel searches rather than one, so the caller effectively has to supply the right vocabulary for each domain separately, which is why the tool exposes `--domain` and `--stack` switches for doing exactly that by hand. And the reasoning engine's job is to filter five independently retrieved sets using rules, not to reason about the request as a whole.

That is a different thing from a model that understands your brief. It is also more predictable, which is an advantage when you want the same input to produce the same design system twice.

Two ways to use it, across nineteen assistants

The installation surface is unusually wide, and it splits into two modes that behave differently.

Skill Mode activates automatically: you describe what you want in natural language, for example asking for a landing page for a SaaS product, and the assistant consults the skill. The README lists the assistants that support it, covering Claude Code, Cursor, Windsurf, Antigravity, Codex CLI, Continue, Gemini CLI, OpenCode, Qoder, CodeBuddy, Droid, KiloCode, Warp and Augment.

Workflow Mode is a slash command you type yourself, and it covers Kiro, GitHub Copilot, Roo Code and KiloCode. The distinction is a property of how each assistant supports plugins rather than of this skill, which is why the list is split rather than unified. There is also a per-client caveat: Trae users are told to switch to the tool's SOLO mode before making a UI request.

Installation runs through a CLI, and the naming here is a trap worth stating precisely. The npm package is `ui-ux-pro-max-cli`, and the command it installs is still `uipro`. The old package, `uipro-cli`, is described as retired and not recommended. So the executable did not change when the package name did, which is the least visible and most annoying kind of rename.

The basic sequence is short:

npm install -g ui-ux-pro-max-cli
cd /path/to/your/project
uipro init --ai claude

The `--ai` switch takes the assistant name and there is an `--ai all` value that covers every detected platform, which is convenient and is also the widest thing you can install into your home directory. Global installation, which writes to a skills directory in your home rather than to the project, has its own flag and its own home per assistant.

The Claude marketplace route exists too, and it is the shortest of the three paths:

/plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill
/plugin install ui-ux-pro-max@ui-ux-pro-max-skill

It installs into one assistant. The CLI exists because it installs into all of them, and because it has the lifecycle commands the marketplace does not.

The CLI is the update channel, not the marketplace

The lifecycle commands are the part of this tool that tells you how updates work, and they point somewhere unusual.

uipro versions
uipro update
uipro init --offline
uipro uninstall

`uipro versions` lists the versions available. `uipro update` refreshes the skill files from the installed CLI package, not from the network and not from the marketplace. That is a deliberate architecture: the npm package is the source of truth for both the installer and the content, and the skill files on disk are a copy that the CLI refreshes when you ask it to.

The consequence is that upgrading the package and upgrading the skill are two different actions, and forgetting the second is the normal failure mode. If you pin `ui-ux-pro-max-cli` in a lockfile or a build image and never re-run `uipro update`, your skills stay at whatever version was current when they were installed. There is no automatic check.

`uipro init --offline` installs from templates bundled inside the package, which is the answer to a locked-down or disconnected environment, and it means the tool is usable in a pipeline that cannot reach npm at install time as long as the package itself was fetched beforehand.

Uninstalling is auto-detecting by default, which is convenient and also slightly risky, since it will find platforms you installed by hand for other reasons. The explicit forms exist for that reason: a named assistant, or the global flag.

There is one prerequisite that surprises people. The search scripts need Python 3, and the tutorial gives the install command for each platform, including `winget` on Windows:

python3 --version
brew install python3
sudo apt update && sudo apt install python3
winget install Python.Python.3.12

A design skill that reaches for a Python interpreter is worth knowing about before you install it into an environment that assumed Node was the only runtime.

Four axes of retrieval, and one that writes files

The advanced commands expose the retrieval directly, which is the right way to develop a feel for what the corpus contains.

python3 .claude/skills/ui-ux-pro-max/scripts/search.py "glassmorphism" --domain style
python3 .claude/skills/ui-ux-pro-max/scripts/search.py "elegant serif" --domain typography
python3 .claude/skills/ui-ux-pro-max/scripts/search.py "form validation" --stack react

So there are four axes: a style domain, a typography domain, a chart domain, and a stack. Each is a different slice of the corpus, and the `--stack` axis is the one that is unusual, because it means the tool holds guidance per technology, twenty-two of them from React and Next.js through SwiftUI, Flutter, JavaFX and WinUI. That is the difference between recommending a design and recommending how to build it.

The design system command is the top of the stack, and it is where the output format becomes your choice:

python3 .claude/skills/ui-ux-pro-max/scripts/search.py "fintech banking" --design-system -f markdown

The ASCII form is for reading in a terminal, the Markdown form is for pasting somewhere.

Then there is the flag that changes the nature of the tool. With `--persist` and a project name, the generated design system is written to disk instead of returned as text, in this structure:

design-system/
├── MASTER.md
└── pages/
    └── dashboard.md

A master file plus one file per page. That is a versionable artefact, and it is the difference between a suggestion that scrolls away and a constraint your team can review in a pull request and disagree with. It also means the output accumulates, so the second invocation with a `--page` flag adds to the tree rather than replacing it, which is the right behaviour for a design system and the wrong behaviour if you were expecting a clean regeneration.

The style count is worth holding onto while reading the output, because it bounds what you can get: sixty-seven named styles, from glassmorphism and claymorphism through minimalism and brutalism to AI-native and spatial interfaces. If the design you want is not in that list, the tool will find you the closest thing in it.

What this thing is not, and the limits of a generated checklist

Two honest limitations, one about the tool and one about the category.

On the tool: its value is that its answers are reproducible and its vocabulary is bounded. Ask twice, get the same design system. That is a genuine property and it is why this kind of corpus is worth having at all, since a model's own priors move between versions. The cost is that you are steering an assistant toward sixty-seven named styles, a hundred and sixty-one colour palettes and fifty-seven font pairings. If your design is none of those, the tool will helpfully give you the nearest one, and the nearest one is a decision you did not make.

On the category: the deliverable includes a pre-delivery UX and accessibility checklist, ninety-nine guidelines covering best practices, anti-patterns and accessibility rules, and a chart-type catalogue for dashboards. None of that is a substitute for testing with users, and a generated checklist is a list of things somebody thought of in advance rather than a list of things your users will do.

There is also a maintenance reality worth stating for this repository specifically. The site says it has been synced with the upstream v2.0 update and lists what changed on the page: the statistics updated to the v2.0 numbers, a new section for the design system generator, a new Claude marketplace install method, the CLI package name corrected, and the supported platforms and stacks extended. That is a well-kept changelog for a hand-maintained mirror, and it is also the entire basis on which the numbers in this article can be trusted.

So the practical advice is unglamorous. Install from upstream, and use this page to decide whether you want to, not to decide whether you have.

Editorial conclusion

Read ui-ux-pro-max-skill-cn if you are installing the upstream skill and want the reasoning in Chinese, and treat it as documentation rather than as the tool: install from npm with `npm install -g ui-ux-pro-max-cli` or through the Claude marketplace, because the command is still `uipro` and the retired `uipro-cli` package will not deliver new skills. Compare the row of statistics against upstream before you trust it, since this site tracks the upstream README by hand. Note that the search scripts need a Python 3 interpreter, and use `--persist` if you want the generated design system to become a reviewable file in your repository rather than a suggestion in a chat window.

Frequently asked questions

Official sources

1. bbylw/ui-ux-pro-max-skill-cn on GitHub
2. Issues
3. License: MIT
4. Project website
5. README