agentleFS
madebymustafamadebymustafa/design-taste/skills/taste-design-system/SKILL.md

taste-design-system

Skill1 starsChanged 3 months agoMITView on GitHub

Builds and audits design systems for agentic workflows: three-tier design tokens (primitive → semantic → component), W3C DTCG token format, spacing and type scales, color ramps, component anatomy, documentation, governance, and adoption metrics. Use when creating tokens, scaffolding a component library, or auditing an existing system for consistency. Stack on the design-taste core skill.

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write tokens as a flat key map (or W3C DTCG JSON if the stack supports it):

    line 28
  2. Distinctive:

    Run the inventory before building anything:

    line 43

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

taste-design-system

A design system is a decision inventory. This skill tells the agent how to take that inventory, structure it, and keep it honest.

1. TOKEN ARCHITECTURE (Three Tiers, Never Two)

  1. Primitive tokens — raw values: --color-ink-900, --space-4, --radius-md. No meaning beyond the value.
  2. Semantic tokens — meaning, not value: --color-surface, --color-text-muted, --color-border, --space-inset-lg. Products consume semantic tokens ONLY.
  3. Component tokens — per-component aliases: --button-bg-primary, --button-radius, --input-border-width. Components never reference primitives directly.

Rule: components → semantic → primitive → value. If any layer is skipped, the system will fork.

2. TOKEN FORMAT (DTCG, Not Freeform)

Write tokens as a flat key map (or W3C DTCG JSON if the stack supports it):

--color-ink-900: #101014;
--color-ink-700: #3B3B40;
--color-accent: #E8540B;
--space-1: 4px; --space-2: 8px; --space-3: 12px; --space-4: 16px; --space-6: 24px; --space-8: 32px; --space-12: 48px; --space-16: 64px;
--radius-sm: 2px; --radius-md: 6px; --radius-lg: 12px;
--font-display: 'Archivo', sans-serif; --font-body: 'Inter', sans-serif; --font-mono: 'JetBrains Mono', monospace;

Naming conventions: kebab-case, noun-first, scale-last (--space-4, --color-ink-900, --radius-md). Never encode intent in names ("fancy-purple"), never use hex colors directly in components.

3. THE INVENTORY (Audit Before You Build)

Run the inventory before building anything:

  1. Colors — list every hex in the codebase. Flag any value not in the token set (>15 unique out-of-scheme colors = system is broken).
  2. Spacing — flag any margin/padding not on the 4px (or 8px) scale.
  3. Type — count distinct font sizes/weights in use. >5 sizes on one page = no hierarchy.
  4. Radius / elevation / borders — collect every value; cluster into 2–4 tokens.
  5. Components — list every component; flag components with >3 variants doing the same job, and any component styling itself from raw values.

Output: a token map, a violation list with counts, and the fix order (primitives first, then semantic mapping, then component cleanup). Never refactor a system in the same pass as auditing it — audit, report, then change.

4. SCALES (Fixed Reference)

  • Spacing: 4/8/12/16/24/32/48/64/96. Gaps below 4px only for hairline borders. Never 5, 7, 10, 14px — unless a deliberate optical override is documented.
  • Type scale (fluid): 12/14/16/18/24/32/48/64 rendered with clamp() where responsive.
  • Color ramps: 9–11 steps from 50→950 in OKLCH (perceptually even), neutrals and at least one accent family. Dark mode = semantic mapping change, not new colors.
  • Elevation: 3 levels max: flat (no shadow), raised (subtle, blur ≥ 16px, opacity ≤ 0.15), floating (blur ≥ 40px, opacity ≤ 0.2). Everything else is a bug.

5. COMPONENT ANATOMY (Every Component Ships With)

  1. One job, one name. If a component has an "and" in its description, split it.
  2. Full state matrix: default, hover, active/pressed, focus-visible, disabled, loading, error. If a state is impossible, document why.
  3. Size variants only where real (sm/md/lg), never color variants pretending to be states.
  4. Responsive behavior and the breakpoint where it changes.
  5. Token references in code — zero raw values inside component styles.

6. DOCUMENTATION & GOVERNANCE

  • One page per component: purpose, anatomy diagram (as code comment), state matrix, usage rules, do/don't examples, accessibility notes.
  • Change process: token changes are additive (new token → deprecate old → remove after one release cycle). Never mutate a token value silently.
  • Ownership: one person (or one skill) holds the token file. Component decisions route through it.
  • Adoption metrics (measure quarterly): % of screens consuming semantic tokens, # of raw-color violations, # of off-scale spacing values, % of components with documented states. The number that matters: violation count trending down.

7. CHECKLIST (Fail the System If…)

  • Any component references a primitive directly.
  • Any UI uses a hex not present in the token set.
  • Any spacing value off the scale without a documented override.
  • Any component missing a disabled or focus-visible state.
  • More than 3 elevation levels in use.
  • Token names encode meaning AND value ("dark-purple" instead of semantic).
  • Two components exist that do the same job with different APIs.

Tokens are the memory of a design system. If the system can't remember, it repeats itself — and so do your pages.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.