Define or refactor a shared frontend UI system before polishing one screen at a time. Use when the user needs token governance, visual-language rules, primitive naming, page-system direction, cross-product consistency, or a design handoff that should stay coherent across landing pages, dashboards, forms, and component libraries. Not for reusable component API architecture, responsive layout-only fixes, accessibility-only remediation, or broad UI critique. Triggers on: design system, design tokens, visual language, component library foundations, UI governance, cross-product consistency, landing page plus dashboard system, primitive naming, token policy.
Define or refactor a shared frontend UI system before polishing one screen at a time. Use when the user needs token governance, visual-language rules, primitive naming, page-system direction, cross-product consistency, or a design handoff that should stay coherent across landing pages, dashboards, forms, and component libraries. Not for reusable component API architecture, responsive layout-only fixes, accessibility-only remediation, or broad UI critique. Triggers on: design system, design tokens, visual language, component library foundations, UI governance, cross-product consistency, landing page plus dashboard system, primitive naming, token policy.
allowed-tools
Bash Read Write Edit Glob Grep
compatibility
Best for frontend and fullstack repos that need reusable system rules across multiple pages or products. Route implementation-only work and specialist remediation to neighboring frontend skills once the system boundary is clear.
Use this skill when the main question is "what shared UI rules should govern this product or product family before we design or refactor individual screens?"
This is the repo's canonical frontend UI-system anchor.
It should:
classify the system problem,
choose the right design-system mode,
route neighboring frontend work out early,
define a compact system packet another human or agent can execute.
Primitive governance mode — when the team needs rules for primitive naming, ownership, contribution boundaries, and promotion from product-local UI into shared primitives
System handoff mode — package already-decided system rules into a concise artifact for design/dev execution
System review mode — assess whether inconsistency is really a design-system problem or should route to a neighboring skill
Quick frame:
Design-system request:
- Surface: marketing site + logged-in dashboard
- Primary mode: cross-surface alignment
- System question: shared tokens and hierarchy, separate page-specific layouts
- Likely route-outs: component API details to ui-component-patterns, responsive verification to responsive-design
Step 2: Define the shared system boundary
Before changing tokens or page direction, decide what the system actually owns.
Keep in design-system when the job is to define:
token scales and naming rules
primitive naming and contribution policy
visual-language principles that should apply across many screens
breakpoint or density policy that multiple surfaces should inherit
motion principles and accessibility baseline that many components depend on
Do not keep the work here when the question becomes:
one reusable component API → ui-component-patterns
one page’s layout breakage → responsive-design
one accessibility failure or remediation plan → web-accessibility
one broad UX/UI critique → web-design-guidelines
Step 3: Make the minimum shared decisions first
Do not start by dumping dozens of example tokens.
For each system request, decide these shared layers first:
Governance — who can add/change tokens or primitives, what must be reviewed, and when work should route to another skill
If a layer is unknown, say so and capture it as an open decision instead of inventing detail.
Step 4: Keep examples subordinate to the rules
Examples are useful only after the shared rules are explicit.
Good example usage:
show one token naming pattern
show one page-system contrast between landing and dashboard surfaces
show one primitive promotion rule
show one motion baseline or accessibility baseline
Bad example usage:
dumping a full palette, TS token file, CSS animation sheet, and JSX screen mock before saying what problem the system is solving
Use this rule:
rules first,
one or two illustrative examples second,
handoff packet last.
Step 5: Route neighboring frontend work out early
Mixed requests are normal. Split them explicitly.
Examples:
if the team asks for button variants, slot structure, and controlled/uncontrolled behavior, keep the system naming/tokens here but route API design to ui-component-patterns
if the team asks how cards collapse on mobile, keep shared breakpoint policy here but route layout adaptation to responsive-design
if the team asks whether icon-only buttons meet keyboard/focus/label requirements, keep baseline a11y expectations here but route remediation to web-accessibility
if the request is “audit this dashboard and tell us what feels off,” route the broad critique to web-design-guidelines
Do not hide unclear ownership by calling everything “design-system work.”
Step 6: Produce the design-system packet
End with a concise artifact that downstream design/dev work can follow.
If the user asks for a terse answer, keep the same sections but compress them into bullets.
Step 7: Only then add a small illustrative example when needed
Use at most one or two examples and keep them obviously subordinate to the packet.
Healthy example shapes:
semantic color tokens (surface/default, surface/emphasis, text/muted) instead of giant fixed palettes
landing-page hero vs dashboard header comparison to show shared hierarchy but different density
primitive promotion rule such as “shared button stays primitive; billing-upsell banner stays product-local”
Step 8: Finish with a boundary sentence
End with a sentence that prevents overlap drift.
Examples:
"This packet defines the shared UI system; component API extraction belongs in ui-component-patterns."
"This packet sets breakpoint and density policy; page-level layout fixes belong in responsive-design."
"This packet sets accessibility baseline expectations; remediation and verification belong in web-accessibility."
Examples
Example 1: shared landing page + dashboard system
Input: "We need one design system for our marketing site and B2B dashboard so tokens, typography, and motion feel related without making both surfaces identical."
Good response shape:
choose cross-surface alignment mode
define shared foundations and surface-specific differences
leave one compact design-system packet
route component-family API design to ui-component-patterns
Example 2: token and primitive governance
Input: "Our team keeps adding random colors and spacing values. We need naming rules and a review policy for shared primitives."
Good response shape:
choose foundations or primitive governance mode
define token/naming rules and review thresholds
keep one or two naming examples only
avoid turning the answer into a full page mock
Example 3: mixed system + responsive request
Input: "Should the dashboard and mobile web app share one breakpoint policy, and how should the layout collapse on small screens?"
Good response shape:
keep shared breakpoint/density policy in design-system
route the actual collapse strategy and verification to responsive-design
make the split explicit in the packet
Example 4: accessibility-heavy follow-up
Input: "We already have tokens, but our forms still fail focus visibility and error-state accessibility."
Good response shape:
note the design-system baseline briefly
route the real remediation and manual verification to web-accessibility
do not absorb the full a11y fix into this skill
Best practices
Decide the mode first so the skill stays bounded.
Define shared rules before examples to avoid overfitting on one mockup.
Keep governance explicit: token changes, primitive promotion, and ownership rules are part of the system.
Use route-outs early so design-system does not steal layout, component-API, accessibility, or broad-audit work.
Leave a compact packet another human or agent can follow.
Keep the alias narrow: frontend-design-system remains a compatibility redirect, not a peer canonical skill.
