UI UX Pro Max

What it actually installs

The repository is a dataset with a search engine in front of it, not a clever prompt. Installation is a CLI: install `ui-ux-pro-max-cli` globally, then run `uipro init --ai <assistant>` inside your project, which writes skill files for whichever tool you use. The README lists a long roster of targets — Claude Code, Cursor, Codex, Copilot, Gemini CLI, Windsurf and others, plus a universal `.agents/skills/` path for anything following that convention — and a Claude Code plugin install as an alternative. The search script needs Python 3 and, per the README, uses only the standard library, installs nothing and makes no network calls.

• 192 industry reasoning rules one per product type — fintech, dental clinic, developer tool, habit tracker — each carrying a recommended page pattern, a style priority, colour and typography mood, and the anti-patterns to avoid for that industry.
• 79 searchable UI styles, 50 of them active the other 29 are supplemental, returned only on an explicit or exact request, and a further 9 deprecated names sit outside the ranking entirely and redirect to a canonical style. Publishing the split rather than one headline number is a more honest catalogue than most of these packs offer.
• 192 colour palettes and 74 font pairings mapped one-to-one against the same product types, with the Google Fonts imports included.
• 34 landing-page patterns held in a separate dataset on purpose, so page structure does not compete with visual style inside the same ranking.
• 119 UX guidelines weighted toward failures that only show up in production: text clipping at browser zoom, chip rows that cannot wrap, badges carrying meaning in colour alone, interrupted animations that settle into the wrong state.

Why a rule pack changes what an assistant builds

Ask a coding assistant for a landing page and you get the average of everything it has seen: a centred hero, three equal cards, a gradient, an emoji standing in for an icon. That is not a limit of the model, it is the absence of a brief. A skill pack narrows the request before generation — "a page for a spa" becomes a named pattern, a palette, a type pairing and a list of things not to do — so the assistant chooses inside constraints instead of reaching for the median.

The pre-delivery checklist is the part worth stealing even if you never install it. Contrast minimums, visible focus states, no emoji as icons, reduced-motion respected, and a look at 375, 768, 1024 and 1440 pixels are precisely the items an AI-built page misses, and a human can check every one of them in two minutes. Our own take on where that lands is in should I use AI to build my website.

What we learned running our own design rules in production

AgentCeres — the AI Growth Officer at agentceres.com — maintains an in-house design skill for the agent that builds and publishes customer landing pages, and we rewrote it in August 2026 for exactly the reason this project exists: the craft layer had been five bullet points, so the output was competent and completely generic. We measured the rewrite the only honest way, running the same model against the same brief with the same image, old skill versus new, and counting the tells:

1. Visible em dashes in the copy ten in the old output, none in the new one.
2. Section eyebrow labels five across six sections, then one.
3. Rows of three equal cards two, then none.
4. Different labels for one call-to-action intent three, then one.
5. Images that survive publication none, then all of them — the old skill produced a hosted image URL that rendered locally and broke on the live page.

That last line is the lesson that generalises past our stack. Published pages are served under a strict content security policy, so a hosted image URL and a web-font stylesheet both worked in the downloaded file and failed silently once live — and nobody noticed for weeks, because the only page anyone had published contained no images. A design rule pack can be entirely correct and still produce a page that breaks, because the rules do not know the envelope the page will be rendered into. Whatever you adopt, verify the output where it will actually live, not where it was made.

Growing what you build with it

A page you are not embarrassed by raises the ceiling on every channel pointing at it, and by itself it moves nothing. The work after the page ships is the ordinary repetitive kind: answer and comparison pages for the questions your buyers actually type, a social cadence you keep to, replies where your category is discussed, and enough measurement to tell which of those did anything.

That is the job AgentCeres does — a roster of specialist agents run by an AI Growth Officer, drafting the work and putting anything outbound in front of a person before it leaves. If a skill pack like this has already solved the design half, the next constraint is usually distribution. Start with how to turn website visitors into signups, and keep the page coherent as it grows with the design system entry.