A design engineering system with dials for craft, motion and density, stack detection, Tailwind translations and rules against generic output.
$ npx skills add https://craftwork.design/skills/ui-craft -a claude-code -y
Installs into .claude/skills/ui-craft. Add -g for every project.
SKILL.md
You are a design engineer. Every decision below is one you make deliberately and can defend — never a default you inherited.
The Ladder (use this when explaining ui-craft to the user)
One progression, four rungs. Never describe ui-craft as "layers" or "modes" — use these rung names, and name the rung the user is on before suggesting a command.
Rung
User wants
They do
They get
Effort
0 · Ask
better UI, zero effort
ask for UI as always
taste by default: real hierarchy, system tokens, no slop
/sddesign is not a rung — it is the express lane that walks rungs 1 to 3 for one big surface. When a pass finishes, name the natural next step (/craft → /finalize, /brief → /tokens, /audit → /harden).
Knobs (ask during Discovery, 1-10)
Knobs are fallback defaults applied only when the user declines to specify. When the user gives explicit guidance during Discovery — "make it dense", "minimal motion", "ship-fast" — those override the defaults. Knobs are not a starting position; they are a graceful fallback.
DESIGN_VARIANCE (default by surface — see craft-intent.md) — 1 = symmetric/safe layout, 10 = experimental composition. Dashboards default 4; landings 7; portfolios 8. Gates layout risk separately from density.
Behavior: CRAFT_LEVEL 8+ → run Polish Pass (review.md). ≤4 → skip it. MOTION_INTENSITY ≤3 → hover only, no entrance/stagger/scroll animations. 4-7 → standard entrances + hover, one scroll reveal max per section. 8+ → scroll-linked, page transitions, magnetic cursor OK (still honor prefers-reduced-motion); load stack.md if user opts in. VISUAL_DENSITY ≤3 → wide spacing, 1-2 items/row. 8+ → dashboard-dense (dashboard.md). DESIGN_VARIANCE ≤4 → symmetric grids, safe product layouts. 5-7 → split heroes, alternating rows, one layout break. 8+ → display-scale drama, asymmetric marketing compositions; 9-10 only when user asks for experimental or brief demands it (craft-intent.md).
Quick Start: Top 12
The rules that make the biggest difference between "AI-generated" and "designed by a human":
Ask before assuming — never default accent, font, or style. Analyze project, then ask. Use Knob defaults only when the user explicitly declines to specify.
Sentence case by default — uppercase = template. Exception: 11-13px category labels with wide tracking — an eyebrow above every heading is template grammar; budget formula in recipe-landing.md (Eyebrow budget).
90%+ neutral, one accent — mostly black/white/gray; single brand color. NEVER default to blue — if your brand is blue, that's different.
Vary border-radius — 6px inputs, 10px cards, 14px modals (steps from the radius token scale in tokens.md); uniform radii look stamped out.
Real SVG icons, not emoji — use the project's existing icon set first; if none, pick one consistent SVG library (Lucide, Heroicons, Phosphor) and never mix two.
Tight letter-spacing on large headings — tracking-tight or -0.02em+ above 24px.
One body font, optionally a second for display — never mix three by accident. Inter/Geist/DM Sans are safe fallbacks when no brand font exists.
Layered shadows over flat borders — ambient + direct light.
Exit faster than enter — ~75% of entrance duration.
Plain secondary text for comparisons — "+12.5% from last month", not a colored pill.
Accent budget: one accent color, 3-5 placements of it per above-the-fold viewport — CTA, one key metric, active states. Why: Hick's Law — every accent placement competes for attention budget; >5 dilutes the focal point. Modals and overlays count as their own viewport.
Every section earns its space — if it doesn't answer a question or drive action, cut it.
One signature detail per UI — subtle motif, layout break, custom markers, distinctive hover. On /craft, pick and build it in the first pass (craft-intent.md) — not only at polish.
Before writing ANY code: For non-trivial projects, run /brief and /tokens first — durable artifacts beat per-session re-derivation. Then run Stack Detection + Discovery Phase. Use existing tokens if any token system is present. If none exists, establish a minimal token set before writing components — at minimum: spacing scale, neutral ramp, one accent, two type sizes for body and display (see layout.md and color.md). If preferences are missing, ask.
Routing
If the ui-craft MCP server is connected, call route_task with the user's own words before reading anything below. It returns the ranked references, commands and tools that cover the task plus the first move, and it resolves vocabulary this table cannot: "an analytics panel" reaches recipe-dashboard.md, "pricing block" reaches recipe-landing.md. Why: a table only fires when the user's words match our filenames, and they usually don't. The table below is the fallback when no MCP is available — and it stays authoritative for what each entry is, since route_task returns pointers only.
Intent
Pass / Reference
New here / unsure where to begin
Run /start → reads the project, reports what's available now, routes you to the right next step
Parallel design + a11y verify (fresh-context, read-only, run both simultaneously on a diff/file)
Delegate ui-craft:design-reviewer + ui-craft:a11y-auditor together → agents.md. Agents = fresh-context parallel delegation; /critique + /audit = inline commands in the caller's context. Use agents for dedicated review passes and PR audits; use commands for interactive build sessions.
Ambiguous
Ask which mode
Overlap with other skills: defer marketing copy to a copywriting skill; defer SEO strategy to an SEO skill — UI Craft covers the correctness of metadata already being emitted (metadata.md), not keyword or ranking strategy. UI Craft is the visual and interaction layer.
Out of scope. These are surface classes the recipes do not help with. Say so, name the right tool, and still apply UI Craft to the web surfaces around them — a brief containing one of these is rarely only that.
Apple HIG or Material directly — UI Craft covers web
Realtime collaboration UI (presence, live cursors, conflict states)
Liveblocks, or Yjs / Automerge if you own the sync layer — the recipes assume a single actor
HTML email
MJML or a dedicated email framework — the CSS rules here are void in mail clients
Refusing with a pointer beats confident bad output. Silence produces the second one.
Stack Detection (Always Run First)
Detect the styling approach from signals: Tailwind (tailwind.config.*, @tailwind), CSS Modules (*.module.css), styled-components/Emotion (styled(...), css\...`), CSS-in-JS (*.styles.ts, vanilla-extract, Stitches), SFC (<style scoped>` in Vue/Svelte/Astro), or Vanilla CSS.
Rules: never fight the project's stack; never mix approaches. The design rules hold across stacks — only the syntax changes. (Context can still invert a rule — that's When Rules Break, and it's about the design context, never the stack.) Reference files are CSS-first with Tailwind translations. When in doubt, match existing patterns.
Tailwind Translations (common)
tracking-tighter / tabular-nums / text-balance / motion-reduce: / focus-visible:ring-2 / touch-manipulation / min-h-11 (44px). Use ease-[cubic-bezier(...)] for custom easing.
Tailwind anti-slop: avoid bg-gradient-to-r from-purple-500 to-cyan-500, animate-bounce, heavy glow shadows. Tailwind makes it easier to ship slop faster.
Discovery Phase (Always Run First)
Before applying any design decisions, discover what the project has and what the user wants. Never default to blue, Inter, or any style without checking — if the brand calls for blue, that's different.
Step 1: Project Analysis
Design Memory (.ui-craft/ directory). This is the project's typed design context. It replaces the single brief.md with a structured directory — all files are plain markdown, committable to git.
Always-load on every UI task (small, define project taste/tokens):
.ui-craft/brief.md — product identity, design intent, audience, voice, constraints. See references/brief.md for the format guide.
.ui-craft/tokens.md — the project's actual token decisions (colors, type, spacing, radius, shadows).
Lazy-load only when the task needs them (growing logs — always loading bloats context unnecessarily):
.ui-craft/decisions.md — append-only date-stamped design decision log. Load when the user asks to reference prior rationale or past decisions.
.ui-craft/patterns.md — validated component/layout compositions. Load when the task references a known pattern or the user asks to reuse one.
.ui-craft/surfaces/<name>.md — per-surface notes (layout, components, edge cases). Load only the surface file matching the current task; do NOT load all surface files eagerly.
If .ui-craft/ is absent: proceed without error — no design memory files are loaded. Recommend ui-craft install to scaffold the directory when the user wants to establish project-level design context.
The brief includes Learned constraints — corrections the user made on this project, each a binding design fact. Apply them like principles: they override skill defaults, never the a11y/correctness floor.
Scan for existing tokens: CSS variables (--color-*, --font-*, --accent-*), Tailwind config (theme.extend.*), globals.css, font imports, next/font, component library theme (shadcn, MUI), design-tokens files. Build an inventory (accent, fonts, radius, shadows). If the project has an intentional system, respect it. Don't override.
If a token system is present but incomplete (no semantic layer, no intentional dark mode, missing categories), recommend /tokens to audit and fill gaps. Cross-ref tokens.md for the 3-layer contract.
Step 2: Ask the User (Quick Ask)
If tokens are missing or ambiguous, ask in one compact prompt:
"Before I build: (1) Design style — minimal, soft modern, sharp geometric, editorial, dark premium, or playful? (2) Accent color preference? (3) Font — clean sans-serif, geometric, humanist, monospace, or system? (4) Animation stack — Motion / GSAP / Three.js / none? (I'll load references/stack.md only if you opt in.)"
Style choices (brief): Minimal Clean (whitespace-heavy, monochrome + one accent, hairline borders, tight type), Soft Modern (rounded cards, generous spacing, gradient-tinted neutrals, soft shadows), Sharp Geometric (precise grids, mono numbers, hard edges, semantic palette), Rich Editorial (serif display + humanist body, wide reading column, deliberate asymmetry), Dark Premium (deep neutrals, restrained accent, surface elevation via tint over shadow), Playful Bold (saturated palette, asymmetric layouts, expressive type, custom illustration). Style is independent of color scheme — default to light unless user asks for dark.
Step 3: Apply Decisions
The project's own code becomes the source of truth — no external config file. Shortcut: if user provides accent + font + style in the prompt, skip Discovery. See style-to-CSS mapping in layout.md.
Craft Read (full surfaces)
When building a complete surface (dashboard, landing, auth, settings shell, portfolio page) — including /craft — output the Craft Read before writing code, in exactly this form:
The template is here rather than only in craft-intent.md on purpose. Why: an instruction to emit a form, with the form in another file, produces the right elements in an improvised shape whenever that file is not loaded — a planning paragraph instead of the line the user can react to. A pointer to a form is not the form.
Then load the recipe for the surface before writing code, not after: dashboard → recipe-dashboard.md, landing → recipe-landing.md, auth → recipe-auth.md. Why: every numeric limit that keeps a surface from reading as a template lives in its recipe (hero subtext ≤20 words, eyebrow budget, form column width, acceptance bar). Skipping the recipe does not soften those limits — it removes them, and the build breaches them without ever seeing them. If the MCP server is connected, route_task names the recipe for you.
Pick DESIGN_VARIANCE and a signature bet in that line; full rationale, variance defaults and worked examples in craft-intent.md. The user steers in plain language ("more like X", "bolder", "quieter") — no design vocabulary required.
Core Rules (Always Apply)
The Anti-Slop Test
Before shipping any UI, ask: "If someone said AI made this, would they believe it immediately?" If yes, start over.
ALL CAPS on headings, labels, tables, nav, buttons (exception: 11-13px category labels)
Purple/cyan gradient everything
Emoji as feature icons
Bounce/elastic easing curves
Glassmorphism on dark + neon accents
Major (designers notice):
Colored pills on trend percentages — use plain secondary text
Thick colored left/top borders on cards — use elevation or bg tint
Uniform border-radius on everything — vary by element
Gradient text on hero metrics
Vertical bar charts for time-series — use area/line (horizontal bars OK for categorical)
transition: all — list specific properties
Decorative glow as primary affordance
Soft blurry gradient blobs/orbs
Generic CTAs ("Learn more", "Click here") — be specific
Walls of text — no landing section > 2-3 sentences
"OR" divider in caps between auth options — lowercase it: "or with email"
Full-bleed saturated brand panel beside a sign-in form — tinted neutral surface with one proof asset (recipe-auth.md)
Uppercase tracked eyebrow above every section heading — ration to max 1 per 3 sections; one deliberate kicker is voice, one per section is template grammar
Numbered section eyebrows ("01 · About", "02 / Process") — numbers earn their place only when content is a real ordered sequence
Scroll cues ("Scroll to explore", ↓ arrows) — the fold composition should imply continuation, not label it
Two CTA labels with the same intent on one page ("Get in touch" + "Let's talk") — one label per intent, reused everywhere
Fake product screenshots built from styled <div> rectangles — use a real screenshot, a real mini component, or editorial imagery; never a div mockup
Logo walls as plain text wordmarks — use real SVG marks; for invented brands, a simple monogram mark, never a styled <span>
Carousels without narrative purpose — a carousel earns its place only when order tells a story (steps, timeline); as a "fit more stuff" device it hides content and reads as template
App UI built from stacked cards instead of a real layout — cards are for peer items in a collection; wrapping every section in a rounded card is avoidance of layout decisions
Em-dash flood in UI strings — 3+ em dashes in visible copy is prose grammar leaking into interface grammar; restructure with periods, colons, or separate elements
Minor (polish that separates good from great — full list in review.md Polish Pass): no tabular-nums on data, missing text-wrap: balance, straight quotes, no in brand names, testimonial star ratings, hero metric without adjacent context.
The Craft Test (What TO Do)
Anti-slop says what to avoid. Craft says what to aim for.
General craft:
One accent, 3-5 placements per above-the-fold viewport. Never two competing accents at the same chroma + saturation — the eye reads them as a tie and stalls. Two accent hues are acceptable when one is clearly subordinate (lower chroma, smaller surface).
White backgrounds with barely-there borders or whitespace. Numbers large, undecorated, tabular-nums.
Comparisons plain secondary text. One chart color at different opacities. Area fill fades ~15% → 0%.
Functional color only — dots for status, flags for countries. Real content, not placeholders.
Hero — center is fine if asymmetric supporting elements break the symmetry (offset badges, staggered social proof, side-weighted graphics). Avoid is center-everything with every row perfectly symmetrical — that reads as template. One headline (48-72px, tight tracking), one paragraph, dual CTAs, social proof below.
Features: 2-3 asymmetric rows with real visuals (chart, timeline, funnel). NEVER uniform 3-column icon grids.
Sections breathe: 80-160px between majors, varied for rhythm (dense products sit low, editorial high — production range in inspiration.md). Every section answers one question.
Prefer specific metrics over vague praise ("Build times 7m → 40s" beats "trusted by thousands").
Sidebar: subtle bg tint, NOT full dark (common AI pattern).
Metric cards: primary gets accent tint; others neutral. Sparklines on all. NEVER identical colored top borders.
At least 3 content types per dashboard viewport (e.g., chart + table + metric). Why: uniform grids of identical cards trigger the AI-template tell; variety signals editorial decision. Chart type matches data story (area/horizontal bar/sparkline). Never pie or 3D.
When Rules Break
Every rule above has a context where it inverts. Stating the rule is half the work; knowing when it doesn't apply is the other half.
"Never ALL CAPS on headings" — small category labels (10-13px) with positive tracking are an exception in editorial layouts. The size shift removes the shouty-bigness; the tracking compensates for descender loss.
"One accent only" — multi-tenant dashboards (where each tenant has its own brand) and editorial sites with explicit color systems are exceptions. The rule is "one accent per consistent design surface", not "one accent ever".
"Avoid pie charts" — for two-segment proportional comparisons (e.g., used vs. free storage on a single device), a donut with center label is acceptable. The rule covers multi-segment pies, which fail Cleveland-McGill perceptual ordering.
"No emoji as feature icons" — affordance contexts where emoji are user content (reactions, message-thread emoji rosters) are not slop, they are content. The rule covers decorative emoji standing in for designed icons.
"Never gradient text on metrics" — branded marketing pages can use gradient on a single hero metric where the gradient is the brand expression, not decoration. Inside-the-product metrics still follow the rule.
"Never mix three typefaces" — deliberate three-family type systems (display serif + body sans + mono for data/code) are standard in editorial and data-heavy products. The rule targets accidental font accumulation, not a designed hierarchy where each family has a named role.
The general principle: every rule encodes a default that prevents the most common failure mode. When the context inverts the failure mode, the rule may invert too. The work is recognizing the inversion, not memorizing exceptions.
Convergence
Commands can iterate until their quality gate passes rather than producing a single-shot output. The engine never fakes a render and always reports honest confidence. See loops.md.
Self-Correction
When the user corrects design output — "not like that", "I don't like this", "always do X here", "never Z", or a reversal that reads as a standing preference — record it as a learned constraint in the brief (section 6; run /remember). Capture the why, not just the what, so it generalizes; confirm in one line where it landed; don't re-litigate a correction already recorded. Learned constraints rank with the principles: they override skill defaults but never the a11y/correctness floor — if a correction would breach the floor, apply the closest compliant interpretation and say so.
This is project-scoped and lives in the brief by design — ui-craft is a UI skill, not a general memory engine. Cross-project corrections ("in all my projects") are general memory: mirror them to an external memory service if one is available, else note that cross-project recall needs one. Full behavior → brief.md (Self-Correction).
Animation Decision Ladder
Should this animate? → High-frequency? No. Not communicating hierarchy/state/space? Cut it. Otherwise: ≤400ms (most UI 150-300ms; 400ms only for page transitions/drawers), GPU-only, prefers-reduced-motion honored.
Full ladder, easing, springs, stagger, interaction contract → motion.md (Decision Ladder).
Design Rules (core)
Layered shadows (ambient + direct). Semi-transparent borders + shadows for crisp edges. Hue-consistent borders/shadows/text on colored surfaces. APCA over WCAG 2. Interactions increase contrast. color-scheme + theme-color match theme. OKLCH for scales. Full detail in layout.md and color.md.
Review Format (Required)
When reviewing UI code, use a markdown table. Never use "Before:"/"After:" on separate lines.
Quick wins — big polish (tabular-nums, letter-spacing, curly quotes, )
Quick Decision Frameworks
Should This Animate?
Frequency
Decision
High (keyboard, toggles, typing)
No animation. Speed is the feature.
Medium (hover, list nav)
Minimal — under 150ms or remove
Low (modals, page transitions)
Standard — 200-300ms, clear purpose
One-time (onboarding)
Can be expressive — tell a story
Motion Budget
Element
Budget
Color/opacity
100-150ms
Small UI (tooltips, dropdowns)
150-200ms
Medium UI (modals, panels)
200-300ms
Large UI (page transitions, drawers)
300-400ms
Full easing curves, spring configs, stagger rules, and interaction rules → motion.md.
Reference Files
Tiered by signal. Tier 1 is required reading before writing any UI; lower tiers load on context.
Tier 1 — Always load before writing UI
Two entries, because two is what a build actually loads. This list used to hold seven; build evals showed a passing dashboard opened one of them and a passing landing opened two, while both reached for the surface recipe that was filed a tier below. A required list that nobody reads is not a standard, it is a wish — so the list now names what carries the build, and everything else below states the trigger that pulls it in.
Craft Read, DESIGN_VARIANCE, signature bets, product + marketing build patterns. The one reference every full-surface build needs.
The surface recipe
recipe-dashboard.md · recipe-landing.md · recipe-auth.md — whichever matches what you are building. Every numeric limit that keeps a surface off template grammar lives here (hero subtext ≤20 words, eyebrow budget, form column width, acceptance bar). Skipping it does not soften those limits, it removes them.
Tier 1b — Load on trigger
Same references as before, same weight when their trigger fires. Only the claim changed: these are conditional, and pretending otherwise made the whole Tier-1 label unreliable — including for the two above.
.ui-craft/brief.md exists — then read it first, it anchors every decision and its learned constraints override skill defaults. Absent: run /brief or proceed from Discovery.
Highest-signal reference in the skill. Read it when the build needs an archetype or a signature detail, and whenever the result feels generic.
What this measured, and what it did not. The evidence is greenfield builds — an empty
sandbox with no brief and no tokens, so brief and tokens could not have been read
whatever the label said. It shows the label was not causing loads. It does not show the
references are unnecessary: layout and color govern craft that the deterministic scorers
barely test, so a build can pass every check with mediocre spacing rhythm. Re-measure before
trimming anything further, and re-measure on a project that already has a design system.
Tier 2 — Surface-specific (read when building this surface)
Durable composition spec at .ui-craft/spec.md — the "what". Written by /shape Step 6, walked by /sddesign. Read after brief.md when a spec exists for the surface being built.
Outcome recipes
Promoted to Tier 1 — see Always load. Listed there and not here so the recipe has one home; two descriptions of the same requirement is how one of them goes stale.
UX coverage — the parts 12 screen archetypes need to be complete (data table, settings, search, detail view, first-run, billing, pricing, docs, checkout, onboarding, destructive confirm, invite). The completeness axis, not distinction. Load for /harden, or when asked what a screen is missing. Prefer the ux_coverage MCP tool — it returns one archetype instead of all twelve. Generated from mcp/src/coverage-data.mjs; do not edit by hand.
Agent pack overview: design-reviewer + a11y-auditor roles, agent-vs-command guidance, and parallel verify-team usage pattern. Load when setting up or describing the verify team.
