agentleFS
ZachArticulateVZachArticulateV/designer-pro-and-seo/skills/design-build/SKILL.md

design-build

Skill10 starsChanged 27 days agoMITView on GitHub

Generate distinctive, production-grade frontend UI code from a design system — starting from an on-brand, accessible, rendered HTML scaffold and elaborating with taste that avoids the generic-AI-slop aesthetic. Trigger when the user says "build the site", "make the UI", "design and code", "create a landing page", "build a component", or wants polished frontend that looks intentional rather than templated.

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 0 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Get the design system: Run design-system-gen if none exists.

    line 37
  2. Very specific:

    Variants: If the user wants 2+, hand the brief + design system to parallel-build.

    line 64
  3. Very specific:

    Pre-delivery: Run design-accessibility and qa-gate before handoff.

    line 66

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “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

design-build

Family: design Status: Stable

Purpose

The frontend code generator — and the payoff of the whole design loop. It takes a design system (from design-system-gen) and renders a real, on-brand starting page via render_page.py (semantic HTML5, the system's actual palette/type/ effects as CSS variables, prefers-reduced-motion fallback, :focus-visible states, contrast-safe buttons), then elaborates it with taste. The differentiator is the anti-slop discipline: distinctive type, color, and interaction instead of the default gradient-purple, rounded-corner look — grounded in the chosen style's characteristics from data/ui-styles.csv.

Triggers

  • "build the site" / "make the UI" / "design and code"
  • "create a landing page" / "build a component"
  • "make this beautiful" / "production-grade frontend"

Inputs

  • Design system (from design-system-gen, or "generate one")
  • Page/component spec (from blast-prompt, or free-text)
  • Stack (HTML+CSS+JS | React | Next.js | Vue | Svelte | Astro)
  • Variants (1 = single build; 2+ = invokes parallel-build)

Steps

  1. Get the design system. Run design-system-gen if none exists; save JSON.
  2. Render the on-brand scaffold — a real, accessible starting point, not a blank page:
    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/design/render_page.py" --design-json ds.json --out page.html   # use `py` on Windows if python3 is absent ($env:CLAUDE_PLUGIN_ROOT in PowerShell)
    (or --product-type/--industry/--keywords to run the engine inline).
  3. Elaborate with taste (anti-slop pass). Replace placeholder copy (use copywriting), shape the sections to the brief, and add distinctive details that match the style's characteristics from data/ui-styles.csv — reject the generic AI look. Honor data/ux-rules.csv (one primary CTA, contrast, etc.). Apply the tell→build-move mapping and the chart-handling guidance in references/design-build/anti-slop.md (it applies the shared references/shared/anti-slop-principles.md rubric at build time). When a section needs a chart, let the data shape pick it deterministically rather than defaulting to a bar or pie:
    python3 "${CLAUDE_PLUGIN_ROOT}/scripts/design/gen_charts.py" --shape time-series-multi --series 4 --colors "#1f77b4,#ff7f0e"   # use `py` on Windows
    It returns a chart type, an A/B/C accessibility grade + fallback, a volume_warning when the series/point count overflows the chart's readable band, and (with --colors) a series-color contrast check reusing the engine's WCAG luminance. The reasoning behind the shape→chart table is in references/design-build/chart-selection.md.
  4. Target the stack. For HTML+CSS+JS, build straight from the scaffold. For React/Vue/etc., translate the scaffold's structure + tokens into components (use design-tokens-emit for the token file).
  5. Variants. If the user wants 2+, hand the brief + design system to parallel-build.
  6. Pre-delivery. Run design-accessibility and qa-gate before handoff.

Outputs

OutputWhat it containsFormatQuality bar (how it is scored)
Built UIProduction frontend elaborated from the rendered on-brand scaffoldHTML/CSS/JS or framework componentsClears the intentionality checklist and passes the swap test in references/design-build/anti-slop.md; keeps the scaffold's a11y guarantees (semantic HTML, focus-visible, reduced-motion, AA contrast)
Build manifestTokens used, sections built, what was elaborated vs. scaffolded, and the one-sentence reason for each brand deviationmd / JSONEvery non-default choice is defensible in one sentence; no unexplained default shipped
Charts / data marks (if any)Chart type matched to data shape, series colors from tokens, text alternativeinline SVG / componentMark fits the data shape; every series/label clears AA; meaning never by hue alone; no fabricated data point

Filed to: the user's project workspace, never into the plugin.

Error Handling

ConditionDetectionBehavior (degrade, never fail)User-facing message
No design system yetno design-system JSON providedrun design-system-gen first, then build"Generated a design system first so tokens stay consistent."
Renderer script unavailablerender_page.py missing or errorsbuild from the design-system JSON by hand using the same tokens/a11y contract"Scaffold renderer unavailable — built from tokens directly; a11y contract preserved."
Copy is placeholderheadline/body still generic on reviewroute to copywriting for concrete, specific copy before handoff"Replaced placeholder copy — generic text is slop, not a shippable draft."
2+ variants requesteduser asks for variantshand brief + design system to parallel-build (shared tokens)"Dispatching N variants via parallel-build; all share one token set."
Internal-tool builduser requests "default styling" modeskip the distinctiveness pass; keep the accessible baseline"Built the plain accessible baseline — skipped the distinctiveness pass as requested."
Bad / empty specnothing to build againstask for the page/component spec; invent nothing"Give me the page or component spec (or say 'landing page') and I'll build it."

Dependencies

  • scripts/design/render_page.py (required) — Python 3.10+, standard library only
  • design-system-gen (tokens), data/ui-styles.csv + data/ux-rules.csv (taste constraints), copywriting (real copy), design-tokens-emit (component token files)
  • scripts/design/gen_charts.py + data/chart-types.csv (optional — deterministic data-shape → chart recommendation with an a11y grade + series-color contrast check; method in references/design-build/chart-selection.md)

Notes

The renderer guarantees a baseline that's already semantic, responsive, and accessible — so the human/taste effort goes into distinctiveness, not boilerplate. For internal tools, the user can request "default styling" mode to skip the distinctiveness pass.

Related: for 2+ variants this skill hands the brief + tokens to parallel-build (Step 5), which runs design-build per variant. That is a prose reference, not a Dependencies edge, so the dependency graph stays acyclic.

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.