generate-ui-from-brand
Skill194 starsChanged 4 months agoMITView on GitHub
Pipeline from a URL or DESIGN.md to tokens, decisions and a UI spec. Use when building UI for an existing brand, auditing a design system or fixing visual inconsistency.
In this repository44 other files it gives its agents.
- algorithmic-color-paletteSkill
- app-shellSkill
- authentic-product-representationSkill
- brand-visual-languageSkill
- button-statesSkill
- clone-websiteSkill
- color-mode-and-themeSkill
- component-family-consistencySkill
- coordinated-data-viewsSkill
- data-display-and-selectionSkill
- dembrandtSkill
- domain-expert-configurationSkill
- elevation-and-depthSkill
- extract-designSkill
- form-designSkill
- gestalt-ui-organisationSkill
- global-toolbar-controlsSkill
- information-architectureSkill
- layout-paradigms-and-consistencySkill
- loading-states-and-perceived-performanceSkill
- marketing-vs-product-systemSkill
- micro-interactionsSkill
- modal-and-overlay-patternsSkill
- modular-scale-typographySkill
- motion-and-storytellingSkill
- nielsen-usability-heuristicsSkill
- notifications-and-recoverySkill
- operational-expert-tool-uiSkill
- performance-and-web-vitalsSkill
- real-world-metaphorsSkill
- repeated-component-alignmentSkill
- responsive-paradigmsSkill
- scroll-areasSkill
- semantic-html-and-seoSkill
- sizing-unitsSkill
- status-colors-and-errorsSkill
- sticky-and-fixed-elementsSkill
- surface-separationSkill
- tab-navigationSkill
- ui-context-and-scopeSkill
- ui-densitySkill
- user-flows-and-guided-pathsSkill
- visual-emphasis-and-hierarchySkill
- wcag-accessibilitySkill
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 71
All MCP extraction tools are async — they return a job_id immediately.
- Distinctive: line 78
Run these in sequence (each extraction launches a browser):
- Distinctive: line 94
Do not use raw extracted values directly.
- Very specific: line 112
Decision rule: if the extracted palette has more than 2 brand colours competing for color-primary, pick the one with highest usage on …
- Distinctive: line 128
Decision rule: if extracted body text is below 16px, override to 16px.
- Distinctive: line 150
All other actions use neutral or outlined styles
- Distinctive: line 158
Run contrast check on normalized tokens:
- Distinctive: line 163
If any fail, darken or lighten the token to meet the threshold.
- Distinctive: line 168
If the brand uses orange as a primary colour, it cannot double as a warning — a distinct amber must be defined for warning states
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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.
Files
generate-ui-from-brand
Type: Pipeline / Orchestrator
Input: URL or existing DESIGN.md
Output: Actionable UI spec with decisions made
Step 1 — Extract
If a URL is provided and Dembrandt MCP is available:
All MCP extraction tools are async — they return a job_id immediately. Poll get_job_status until status is "completed", then read result.
{ job_id } = get_design_tokens({ url })
{ result } = get_job_status({ job_id }) // repeat until status === "completed"Run these in sequence (each extraction launches a browser):
get_design_tokens, get_color_palette, get_typography, get_component_styles, get_spacingIf Dembrandt MCP is not available, run CLI:
npx dembrandt <url> --design-md --crawl 3If DESIGN.md already exists: parse it directly — skip extraction.
Step 2 — Normalize Tokens
Do not use raw extracted values directly. Map them to a semantic system first.
Colours
Identify the role of each extracted colour:
| Role | Token | How to identify |
color-primary | Main brand colour | Used on primary buttons, links, key interactive elements |
color-secondary | Supporting brand colour | Used on secondary actions, accents |
color-surface | Background | Page or card background |
color-surface-raised | Elevated surface | Cards, panels, modals |
color-border | Border / divider | Input borders, separators |
color-text | Primary text | Body copy |
color-text-secondary | Secondary text | Labels, metadata, captions |
color-error | Error state | Red — do not assign to any other role |
color-warning | Warning state | Orange/amber — do not assign to any other role |
color-success | Success state | Green — do not assign to any other role |
Decision rule: if the extracted palette has more than 2 brand colours competing for color-primary, pick the one with highest usage on interactive elements.
Typography
Map extracted sizes to a scale. Verify ratio coherence — if sizes do not follow a consistent ratio, round them to the nearest modular scale step (base 16px, ratio 1.25 recommended).
| Token | Min size | Role |
text-base | 16px | Body copy — never below 16px |
text-sm | 14px | Labels, captions — use sparingly |
text-lg | 20px | Lead paragraph |
text-h4 | 25px | Section subheading |
text-h3 | 31px | Section heading |
text-h2 | 39px | Page subheading |
text-h1 | 49px | Page heading |
text-display | 61px | Hero / landing only |
Decision rule: if extracted body text is below 16px, override to 16px.
Spacing
Identify the base spacing unit from the most common small margin/padding value. Derive a scale:
base = extracted smallest recurring value (usually 4px or 8px)
scale = base × 1, 2, 3, 4, 6, 8, 12, 16Border Radius
Extract the most common radius value used on interactive elements (buttons, inputs). This becomes --radius-button — applied uniformly to all buttons regardless of variant.
Step 3 — Apply UX Decisions
With normalized tokens, make the following decisions explicitly. Do not leave these open:
Visual Hierarchy
- Identify the single primary action for the UI being built
- Assign
color-primaryto that action only - All other actions use neutral or outlined styles
- Apply
cursor: pointerto all interactive elements
Gestalt Grouping
- Define spacing between related elements (tight:
space-2) and between groups (loose:space-6orspace-8) - Confirm that related controls will be co-located in the layout
Accessibility (WCAG 2.2 AA)
Run contrast check on normalized tokens:
color-textoncolor-surface: must be ≥ 4.5:1color-text-secondaryoncolor-surface: must be ≥ 4.5:1color-primaryon white/surface (button label): must be ≥ 4.5:1
If any fail, darken or lighten the token to meet the threshold. Document the adjustment.
Error / Status Colours
- Confirm
color-erroris red and used only for errors - Confirm
color-warningis orange/amber and used only for warnings - If the brand uses orange as a primary colour, it cannot double as a warning — a distinct amber must be defined for warning states
Step 4 — Output UI Spec
Produce a concrete, copy-pasteable output. Choose the format that fits the request:
Design Token File (CSS)
:root {
/* Colours */
--color-primary: <value>;
--color-secondary: <value>;
--color-surface: <value>;
--color-surface-raised: <value>;
--color-border: <value>;
--color-text: <value>;
--color-text-secondary: <value>;
--color-error: <value>;
--color-warning: <value>;
--color-success: <value>;
/* Typography */
--font-sans: <extracted font family>;
--text-base: 1rem;
--text-sm: 0.875rem;
--text-lg: 1.25rem;
--text-h4: 1.563rem;
--text-h3: 1.938rem;
--text-h2: 2.438rem;
--text-h1: 3.063rem;
/* Spacing */
--space-1: <base>px;
--space-2: <base×2>px;
--space-4: <base×4>px;
--space-6: <base×6>px;
--space-8: <base×8>px;
/* Borders */
--radius-button: <extracted>px;
--radius-card: <extracted or radius-button + 2>px;
/* Elevation */
--shadow-card: 0 1px 3px rgba(0,0,0,.10), 0 1px 2px rgba(0,0,0,.06);
--shadow-modal: 0 10px 15px rgba(0,0,0,.08), 0 4px 6px rgba(0,0,0,.05);
}UX Audit (if auditing an existing UI)
CONTRAST ISSUES
- color-text-secondary on color-surface: X.X:1 (required 4.5:1) → fix: darken to #...
HIERARCHY ISSUES
- 3 primary buttons on same screen → reduce to 1 primary, 2 secondary
CONSISTENCY ISSUES
- Button radius varies (4px, 8px, 12px) → standardize to --radius-button: 8px
MISSING STATES
- No disabled state defined for inputs
- No error state for form fields
FONT SIZE VIOLATIONS
- Caption text at 12px → minimum 14pxComponent Structure (if generating a layout)
Page layout:
Header (fixed, 56px)
Logo | Nav | [Global controls: small type]
Main
Hero section
H1 (display scale) + lead text + primary CTA (color-primary, full)
Feature grid (3-col)
Card (shadow-card, radius-card) × 3
Icon + H4 + body text
Footer
Links (text-sm, color-text-secondary)
Language/currency selector (text-sm)
Primary action: CTA button in hero
Secondary actions: nav links, card CTAs (outlined)
Error states: inline, adjacent to field, red text + iconRunning This Across Many Brands (Token Architecture & Governance)
Building several brands from one pipeline, don't treat each as a fresh start. One system, many skins: every brand generates from the same semantic tokens; only primitives (colour, type, radius) change per brand. A fix propagates everywhere.
Consolidate periodically — token systems drift. Across many brands and over time, tokens fork, one-off values creep in, and the systems diverge. Schedule a recurring consolidation pass:
- Re-extract and compare the live sites (use Dembrandt's extract + drift/compute-drift tooling, plus plain visual inspection and benchmarking against each other and against current design trends and best practice).
- Fold divergences back into the shared semantic layer where they should be common; keep genuinely brand-specific values as primitive overrides only.
- Feed in a point of view. Consolidation isn't just mechanical de-duplication — bring UX / visual-design opinion and a clear direction for where each product should go, not just where it is.
When design eras conflict, recency is the tiebreaker. A product built over years carries pages from different design generations. The newest pages and components are the best available evidence of current design intent — migrate old toward new; never average the eras into a compromise style. Recency is a default, not a verdict: the newest surface can be an unreviewed one-off. Confirm with the user before promoting a style to canonical or marking an old one deprecated.
Deprecate on touch, not big-bang. Keep a short list of deprecated styles — the old radius, the old shadow, the retired button variant. When work already touches an old-generation page, lift it to the current style in the same pass. Every touch moves the product one page closer to one system, with no rewrite project on the roadmap.
Two or three occurrences make a pattern. When the same visual treatment appears independently in 2–3 places, it is no longer a coincidence — name it, tokenise it, and make it available everywhere. Promoting it is cheaper than a fourth hand-rolled copy. Once promoted, the pattern belongs to the design language: new components may use it as-is or adapt it, as long as the adaptation stays recognisably true to the original.
Track feature usage and deprecate the dead weight. The same discipline applies to features, not just tokens: instrument what actually gets used, and deprecate the features/components with little real usage rather than maintaining them forever. A shared system stays healthy only if it's pruned — every unused component is drift waiting to happen and a cost on every future change.
Audit Checklist Before Handoff
- All tokens named semantically, not by value (
color-primarynotcolor-blue-600) - Body text ≥ 16px everywhere
- Contrast ratios verified for all text/background combinations
- One primary button per view
- All buttons share
--radius-button cursor: pointeron all interactive elements- Error colour reserved exclusively for errors
- Warning colour (orange) reserved exclusively for warnings
- Spacing derived from a single base unit
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
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.

