First Website Kit Alternatives (2026)

A drop-in rulebook that makes your AI design first: tokens before components, a clear primary action per screen, real states, honest verification before "done" — plus a /ui-review prompt for your worst screen. Free. Your code works. Now make it look like it. You already ship real software with Claude Code, Codex, Cursor, Gemini CLI, Copilot, Windsurf — whichever you use. The product works. And the UI looks like every other AI-generated app: default shadcn, a gradient hero, three cards with emoji, forty controls on one screen, no empty states. That is not a taste problem. It's a missing design system. Nobody told the model what "good" looks like in your repo, so it reached for defaults. Hand it the rules and it stops. What this is UI Kit for Builders is a folder you drop into your repo: a rulebook your AI loads at the start of every session, shipped under every filename AI tools look for — AGENTS.md (Codex, Cursor, most tools), CLAUDE.md (Claude), GEMINI.md (Gemini CLI), .github/copilot-instructions.md (Copilot), .windsurf/rules/ (Windsurf) — identical bodies, so it works with whichever one you use — plus a design-system template, a review prompt, and a license-checked repo list. It makes your AI design first: Tokens before components. If your repo already has a token system it builds inside it — never re-themes a working app on a one-screen task. If it doesn't, the first UI turn creates a one-page DESIGN.md (template included) and the token block where your stack reads it (Tailwind v4 @theme / v3 config + shadcn variables): type scale, 8px spacing rhythm, one gray + one accent scale, radius and shadow scales, a motion language, locked primitives. Everything after composes those. The banned list — 37 starter smells, each with the fix. Default shadcn, gradient heroes, emoji feature cards, glass blobs, centered-everything, feature dumping, control overload, system font stacks, cards inside cards, hand-typed hex, placeholder copy, one-off styles, more than one hero effect, linear animation, missing states, dark mode by inversion, dashboards of identical cards… The UX floor, per screen. A clear primary task; hierarchy that survives the squint test; mobile at 375px; loading / empty / error states; forms with labels, inline validation, focus; keyboard and focus rings; WCAG AA contrast; reduced motion; clean console; unique meta. The AI runs it itself and fixes what fails. Motion + effects doctrine. One memorable moment, then restraint. Spring/ease-out, a small CSS-only micro-interaction floor (more only where the product's moment calls for it), and a straight answer on when 3D is warranted (rarely) and how to do it right when it is — the same rules we use on iScaleLabs' own sites. The stack and how to keep it from looking default. Tailwind v4, shadcn/ui (always themed), Radix Colors, Motion, Lucide, Magic UI for one or two accents, R3F for 3D — with the license caveats that bite. Honest verification before "done". No self-graded score: it names any banned row still failing, what it verified (build, console, viewports, live URL) and — plainly — what it could not. /ui-review. Paste it with a screenshot or URL of your worst screen: verdict, then the top 5 fixes with exact code changes. Tool-agnostic. The doctrine is framework-agnostic; the concrete defaults are React / Tailwind / shadcn — on another stack, apply the same rules with your stack's equivalents. No install, no dependency. How to use it (3 steps) Get it from GitHub: https://github.com/iScale-Builders/ui-kit-for-builders — Code → Download ZIP (or grab the latest release, or git clone) — then copy its files into your repo root — the rulebook is already there under your tool's filename (or paste it into your existing instructions file under a ## UI Kit heading). Open the repo in your AI tool as usual. Either start a task and watch it create DESIGN.md + tokens before the first component — or run /ui-review on your worst screen and take the top 5 fixes. Who it's for Builders who already ship with AI and want their product to look as good as it works Anyone whose app "does the thing" but reads as a template Teams that want one design floor every AI in the repo obeys What's in the folder AGENTS.md — the rulebook (~500 lines), read automatically by most AI coding tools CLAUDE.md · GEMINI.md · .github/copilot-instructions.md · .windsurf/rules/ — the same rules under the filenames the other tools look for DESIGN.template.md — the one-page design-system file your AI fills in UI-REVIEW-PROMPT.md — the standalone review prompt, paste-ready REPOS.md — the curated open-source list, every entry license-verified, each with "how to use it so it doesn't look default" VERSION.txt · LICENSE (MIT) Read the rulebook first if you like: https://github.com/iScale-Builders/ui-kit-for-builders/blob/main/AGENTS.md — and star/fork/open issues there; that's where the kit lives and evolves. Current version: see the repo's Releases page (https://github.com/iScale-Builders/ui-kit-for-builders/releases); the version is at the top of the README and in VERSION.txt. Tell us how it went Run /ui-review on a screen, apply the top 5, and post the before/after here. If a rule produced worse UI in practice, that's the most useful feedback there is.