Design — site-aware frontend component builder

You are running the design skill. Your job: create production-grade frontend components that match the target site's existing design system. You wrap the official frontend-design skill with pre-flight context loading and post-processing.

Architecture

Pre-flight: Read the site's CSS variables, fonts, colors, and existing component patterns to build a design context brief.

Delegation: Pass the enriched context to the frontend-design skill for actual code generation.

Post-processing: Verify generated code uses existing CSS variables, replace hardcoded values where possible, validate accessibility.

Step-by-step

Step 0 — Determine site root and mode

Parse $ARGUMENTS:

• If starts with figma → Mode A (Figma-to-code), extract URL
• If starts with like → Mode B (reference-inspired), extract URL
• Otherwise → Mode C (from description)

Site root: --root <path> argument from $ARGUMENTS, or default to current working directory. Convention: site repos live at ~/Sites/<name>/.

Step 1 — Pre-flight: build design context brief

Read up to 3 CSS files from the site root. Extract:

1. CSS custom properties (--variable-name: value)
2. Font stack (font-family declarations)
3. Color palette (from CSS variables)
4. Spacing scale (from CSS variables)
5. Existing component patterns (read 1-2 HTML files to understand button, card, form styles)

Compile into a design context brief — a structured summary like:

DESIGN SYSTEM BRIEF:
- Colors: --electric: #6C5CE7, --neon: #00F5D4, --warm: #FF6B6B, --ink: #1A1A2E
- Fonts: Montserrat (headings), Nunito (body), JetBrains Mono (code)
- Spacing: --s1: 0.5rem, --s2: 1rem
- Patterns: .btn (rounded, uppercase), .card (border-radius, shadow), .fade-up (scroll animation)

Step 2 — Mode-specific preparation

Mode A (Figma):

1. Parse fileKey and nodeId from the Figma URL
2. Call mcp__claude_ai_Figma__get_design_context with clientLanguages: "html,css,javascript"
3. Present to user: design context brief + which CSS vars can be reused + which are new

Mode B (Reference URL):

1. Use the scrape_site.py tool or WebFetch to get the reference page
2. Spawn an Opus Explore subagent to analyze: layout patterns, color usage, typography, interactions
3. Present to user: design brief (what to adopt from the reference, what to adapt to our brand)

Mode C (Description):

1. Present to user: proposed design approach before generating

APPROVAL GATE: In all modes, wait for user approval before generating code.

Step 3 — Generate code via the frontend-design skill

Delegate code generation to the frontend-design skill (Skill tool, skill: frontend-design). The aesthetic principles — typography, color, motion, spatial composition, anti-generic-AI guidelines — live in frontend-design. Do not duplicate them here.

Compose the prompt you pass to frontend-design as a single brief covering:

1. The user's request (Mode A: Figma context blob; Mode B: reference brief from the Explore subagent; Mode C: free-form description).
2. Site-awareness constraints that frontend-design must respect:
   • Brand variables — list the site's CSS custom properties verbatim from the design context brief and instruct the skill to use var(--name) instead of fresh hex/font values whenever a matching variable exists.
   • Output target — vanilla HTML/CSS/JS unless the pre-flight scan detected an existing framework (React, Vue, Astro, etc.); name the framework if so.
   • File structure — how the site organizes pages and shared CSS (one HTML file per page, single shared stylesheet, etc.) so the generated code drops in cleanly.
3. Aesthetic latitude — frontend-design is encouraged to commit to a bold direction; the only hard constraints are the brand variables above. Do not pre-narrow its aesthetic choices.

Capture the generated code from the frontend-design result. Do not write it to disk yet — Step 4 may rewrite hardcoded values.

Step 4 — Post-process

After generating code:

1. Scan for hardcoded hex colors that match existing CSS variables → suggest replacing with var(--name)
2. Scan for hardcoded font-family that matches the site's stack → suggest var(--font-*)
3. Flag any new CSS variables introduced → add them to the shared CSS file with comments
4. Run python3 tools/site_audit.py --root "<site_root>" --mechanical-only --checks accessibility_mechanical,link_integrity on the output

Report: what was generated, which files were created/modified, any CSS variables added.

Important constraints

1. Brand consistency first. Always use existing CSS variables before introducing new ones.
2. Approval gate is mandatory. Show the approach before generating code.
3. Plain HTML/CSS/JS by default. Only use a framework if the site already has one.
4. Post-process is mandatory. Always check for hardcoded values and accessibility.

See also

• workflows/figma_to_code.md — the SOP
• tools/site_audit.py — post-process validation