Communitygithub.com

whoamnesia/design-system

astudio design system and design code: three-layer design tokens (primitive → semantic → component) in W3C DTCG 2025.10 format, CSS custom properties with light/dark themes, a Tailwind v4 theme, fluid type and spacing scales, component specs with every state (buttons, inputs, cards, nav, dialogs, tables, badges), and DESIGN.md - generated from a brand kit or reverse-engineered from an existing codebase or live site. Use when the user asks for design tokens, a token architecture, CSS variables, a Tailwind theme, theming/dark mode at system level, component specifications, "дизайн-код", "дизайн-система", "токены", "UI kit", DESIGN.md, or wants hard-coded values replaced with tokens. Part of the astudio-aio pipeline (after brand, before the site build).

What is design-system?

design-system is a Claude Code agent skill that astudio design system and design code: three-layer design tokens (primitive → semantic → component) in W3C DTCG 2025.10 format, CSS custom properties with light/dark themes, a Tailwind v4 theme, fluid type and spacing scales, component specs with every state (buttons, inputs, cards, nav, dialogs, tables, badges), and DESIGN.md - generated from a brand kit or reverse-engineered from an existing codebase or live site. Use when the user asks for design tokens, a token architecture, CSS variables, a Tailwind theme, theming/dark mode at system level, component specifications, "дизайн-код", "дизайн-система", "токены", "UI kit", DESIGN.md, or wants hard-coded values replaced with tokens. Part of the astudio-aio pipeline (after brand, before the site build).

Works with✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/whoamnesia/site-design/tree/HEAD/skills/design-system

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

astudio · Design system

Role: Senior design-systems designer at the digital studio astudio.one (15+ years) - every request is a well-paid commission that needs a unique, modern, confident result; the bar (references, wow, richness, stock-only photos, no AI images) is in <root>/reference/studio-bar.md.

The design code turns brand decisions into numbers every screen obeys. Rule: components reference semantic tokens only; semantic tokens reference primitives; primitives hold raw values. Change the semantic layer and the whole product re-themes.

Shared: <root> = two levels above this skill's base dir (${CLAUDE_PLUGIN_ROOT}). Tools: palette.py, tokens.py, search.py, render.cjs, lint.py (<root>/reference/tools.md). Local references: references/tokens.md (architecture, naming, DTCG format), references/components.md (specs and states), references/extract.md (from code or URL), references/design-md.md.

Modes

  1. Build from a brand kit (pipeline): palette.json + faces + shape/motion decisions → tokens → component specs → DESIGN.md → a rendered specimen page.
  2. Extract from an existing project (references/extract.md): trace the real tokens, themes, primitives and variants the product renders; normalise to the semantic schema; write DESIGN.md describing what governs the product (not every value that happens to exist).
  3. Audit / migrate: find hard-coded colours, sizes, radii, shadows and durations (lint.py, grep), map them to tokens, propose a migration plan in order of reach, then replace.

Build workflow

  1. Inputs: brand/palette.json (from palette.py system --json), faces (verified, with Cyrillic if needed), radius logic, depth policy, motion tokens, density. Missing? Get them from astudio-aio:brand or decide with the direction contract.
  2. Spec: python3 <root>/scripts/tokens.py init > brand/brand.spec.json; edit name, faces, type min/max sizes and ratios (≈1.2 dense UI, 1.25 product, 1.333+ expressive), spacing base (4), radius scale, shadows (what floats), motion.
  3. Build: python3 <root>/scripts/tokens.py build brand/brand.spec.json --out design-tokens → DTCG <brand>.tokens.json, tokens.css (light + dark via prefers-color-scheme and [data-theme]), and a self-contained Tailwind v4 entry tailwind-theme.css (use it instead of tokens.css in Tailwind projects).
  4. Component layer: define component tokens and specs for the components the product needs (references/components.md): variants, sizes, every state (default, hover, focus-visible, active, disabled, loading, error, selected), anatomy, spacing, accessibility notes. Add component tokens only where a component must be themable independently (--button-radius: var(--radius-md)), not for everything.
  5. Specimen page: build design-tokens/specimen.html (or a Storybook/page in their stack) showing colour roles with contrast values, the type scale at real sizes, spacing, radii, elevation, motion tokens and every component state in light and dark. Render at 1440 and 390, look, fix. This proves the system.
  6. DESIGN.md (references/design-md.md): the human-and-agent-readable design code: atmosphere, roles with hexes, type rules, components, layout, motion, imagery, voice, do/don't.
  7. Wire it: import tokens in the project entry (Next: app/globals.css; Vite: src/index.css; static: <link>), map Tailwind/shadcn variables if used (ui-styling skill), and run lint.py + grep for raw hex in components.

Rules

  • Semantic names describe purpose (--color-ink, --color-surface, --space-section), not value (--blue-600 in a component is a bug). Token names may evoke the product's world (--ink, --parchment) - someone reading only the tokens should guess what product this is.
  • One spacing base (4px); one radius logic (concentric nesting); one shadow policy (border OR tone step OR shadow per layer); one set of motion curves.
  • Dark theme is designed, not inverted: lighter surface steps for depth, adjusted chroma, every pair re-checked.
  • Fluid type with clamp() and rem (zoomable); nothing below 12px; Operate screens use a tighter fixed UI scale.
  • Every interactive component specifies focus-visible and disabled states; targets ≥ 24px (44px on touch).
  • Document why, not only what: each rule in DESIGN.md has its reason so later work doesn't "fix" it away.

Individual skills in this repo

This repo contains 4 individual skills — each has its own dedicated page.

whoamnesia/brand

astudio brand kit and brand identity: positioning, messaging, voice and tone, visual identity (logo usage, colour system, typography, imagery, iconography, shape language), the design code handoff and a rendered HTML brand book. Use when the user asks for a brand kit, brand book, brand guidelines, identity system, фирменный стиль, брендбук, бренд-кит, дизайн-код бренда, tone of voice, messaging framework, or wants to audit whether materials are on-brand, extend an incomplete brand for screens, or apply a brand to a product. Usually runs after astudio-aio:logotype and before astudio-aio:design-system; part of the astudio-aio pipeline.

whoamnesia/cta-banner

astudio CTA banners and ad creatives built for clicks and conversions: offer and audience analysis (awareness stage), scroll-stopping hooks, angle variations (pain, desire, proof, identity, contrarian, urgency), headline and CTA copy formulas, conversion-focused composition (one focal point, one action, contrast, visual flow to the button), platform specs and safe zones for Meta/Instagram, VK, Telegram, TikTok, YouTube, Google Display, Yandex Direct, and an A/B test matrix - all on the brand's design code, exported at exact sizes. Use when the user asks for an ad, ad creative, promo banner, "креатив для рекламы", "баннер с высоким CTR", "рекламный баннер", "креативы для таргета", "продающий баннер", "креатив под A/B тест", or a banner whose goal is a click or sale. Builds on astudio-aio:banner-design.

whoamnesia/logotype

astudio logo design: from brief to a professional, ownable logo built as clean hand-written SVG - word map, 8-12 concepts across mark types (wordmark, lettermark, symbol, abstract, combination, emblem, mascot), three built concepts tested at 16px-billboard, in one colour, reversed, mirrored and in real contexts, then the full logo kit (lockups, small-size cut, one-colour/reversed files, favicon, app and web icons, usage guide). Use whenever the user asks for a logo, logotype, wordmark, monogram, brand mark, emblem, favicon or app icon, a logo redesign or refresh, or critique of a logo - "сделай логотип", "нужен логотип для кофейни", "лого для бренда", "фирменный знак", "иконка приложения из логотипа". Pulls in brand, banner-design or other astudio sub-skills when the request extends to a brand kit or applications.

whoamnesia/ui-ux

astudio UI/UX design intelligence for any interface task - designing, building, reviewing, auditing, critiquing, polishing or fixing pages, screens and components: hierarchy, layout, typography, colour, spacing, interaction, forms, navigation, states, accessibility (WCAG 2.2), responsive behaviour, motion and UX copy, backed by a searchable database (90 styles, 190+ product types with palettes and reasoning, 74 font pairings, 1900 Google Fonts with Cyrillic filter, 119 UX rules, landing patterns, chart guide, icons, GSAP presets, 22 stack guideline sets). Use when the user asks to improve or review a UI, "make it look better/premium/less generic/less AI", audit accessibility or UX, fix a layout, choose a style/palette/fonts for a product, "проверь интерфейс", "улучши UI", "сделай красивее", "аудит UX", or for any focused UI decision outside the full astudio-aio pipeline.

Related Skills