Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Produce a world-class single-page editorial landing site in the Atelier Zero visual language (Monocle / Apartamento / Études editorial collage) — the same aesthetic Open Design uses for its own marketing surface. The agent fills a typed `inputs.json` from a brand brief, optionally generates 16 collage assets via gpt-image-2, then runs a pure-function composer that emits a self-contained HTML file; a separate path can mirror the Astro marketing site in `apps/landing-page/`. Drop-in scroll-reveal
.claude/skills/nexu-io-open-design-landing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 101% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 172% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 118% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 55% | 0% |
Build a single-page editorial landing site (or a slide deck — see the sibling open-design-landing-deck skill) in the Atelier Zero design system: warm-paper background, Inter Tight + Playfair Display, italic serif emphasis spans, dotted hairline rules, coral terminating dots, scroll-reveal motion, and 16 surreal collage plates.
This is the canonical OpenDesign marketing-page recipe — the example output is the very page you see at open-design.
The skill is fully parameterized. The agent fills one typed inputs.json from the user's brief; the composer turns that JSON + the canonical styles.css into a deployable artifact.
textinputs.json + styles.css 16 image slots │ │ └──────────► scripts/compose.ts ◄────────────┘ │ ▼ <out>/index.html (self-contained) <out>/assets/ (PNG or SVG)
A single HTML file with all of:
sticky nav with live GitHub star count.
hero (with 3 stat rings + 4-step index), about, capabilities (4 cards), labs (5 cards + filter pills + progress bar), method (4 steps with thumbnails), selected work (dark slab + 2 tilted cards), testimonial (pull quote + 5 partner glyphs), CTA (ribbon + email pill).
prefers-reduced-motion).
Run these four steps in order. The agent should complete each step before moving on, and prefer asking the user a focused question over inventing copy.
Use AskQuestion (or the equivalent in your UI) to collect the brand brief in chunks; do not dump the entire schema.ts on the user. Map their answers into inputs.json matching the typed shape.
The eight question groups, in order:
| Group | Schema fields | Min answers | Notes | | :---- | :------------------------------------------------------ | :---------- | :--------------------------------------- | | 1 | brand.{name,mark,tagline,description,location} | 5 | Mark = single glyph (Ø, ▲, ★…) | | 2 | brand.{license,version,year,primary_url,contact_email}| 4 | URL is required; license defaults Apache-2.0 | | 3 | nav[] (up to 5) | 3 | Optional count badges | | 4 | hero.{label,headline,lead,primary,secondary,stats} | All | Headline as MixedText (sans+em+dot) | | 5 | about + capabilities.cards[4] | All | 4 cards × {num,tag,title,body} | | 6 | labs.cards[5] + method.steps[4] | All | Both grids fixed-arity | | 7 | work.cards[2] + testimonial | All | 5 partner glyphs as inline SVG path data | | 8 | cta + footer.{columns[4],mega} | All | Mega kicker is a MixedText like the headlines |
Open inputs.example.json for a complete worked example (OpenDesign itself).
| Strategy | When to choose | Cost / latency | | :---------------- | :------------------------------------------------------ | :-------------------- | | placeholder | First pass. Demo. Slide internal. No image budget yet. | $0, <1s | | generate | Final delivery. Brand wants original collages. | ~$0.40, ~6 min | | bring-your-own | User has art direction PNGs. Drop them at assets_path.| $0, 0s |
Set inputs.imagery.strategy accordingly.
placeholder — frame modebashnpx tsx scripts/placeholder.ts <out>/assets/
Writes 16 .svg files (with .png aliases for compatibility) into <out>/assets/. Each placeholder shows the slot id, ratio, pixel dimensions, and the prompt hint from image-manifest.json. The composer's <img src='./assets/hero.png'> etc. just work.
generate — gpt-image-2 modebashFAL_KEY=... npx tsx scripts/imagegen.ts <inputs.json> --out=<out>/assets/
Calls fal.ai's openai/gpt-image-2 synchronous endpoint per slot. Composes prompts as: style anchor (paper-collage editorial system) + brand variables (name / nav / headline / italic emphasis pulled from inputs.json) + per-slot composition (e.g. cropped plaster head + tree growing through arch). Skips slots whose target file already exists; pass --force to re-render.
Without FAL_KEY, the script prints the prompts so the operator can route them through the /gpt-image-fal slash-command skill manually.
bring-your-ownDrop 16 PNGs matching assets/image-manifest.json filenames at inputs.imagery.assets_path. Done.
bashnpx tsx scripts/compose.ts <inputs.json> <out>/index.html
The composer reads inputs.json and ../styles.css, then writes one self-contained HTML file. The page includes:
data-reveal attributes for staggeredscroll motion.
apps/landing-page/app/_components/reveal-root.tsx).
header.tsx).brand.primary_url).For deployable production output, fork the apps/landing-page/ package: copy it into your workspace, align app/page.tsx with content from your inputs.json, and copy your <out>/assets/*.png into the paths expected by app/image-assets.ts / R2 URLs. Build with pnpm --filter @open-design/landing-page build for a static out/ export ready for any CDN.
> A future iteration may bundle a composer that emits the full > apps/landing-page/ tree from inputs.json in one command. Until > then, fork-and-edit is the supported path.
Before marking done, the agent must verify:
<out>/index.html opens in a browser without console errors.display h1/h2 end.prefers-reduced-motion: reduce (DevTools → Rendering) disablestransitions cleanly.
hero (CLS < 0.05).
textdesign-templates/open-design-landing/ ├── SKILL.md # this contract ├── README.md # quick-start ├── schema.ts # typed inputs (single source of truth) ├── styles.css # Atelier Zero stylesheet (single source of truth) ├── inputs.example.json # OpenDesign as the worked example ├── example.html # canonical rendering (regenerated from inputs.example.json) ├── scripts/ │ ├── compose.ts # inputs.json + styles.css → index.html │ ├── imagegen.ts # gpt-image-2 wrapper (fal.ai) │ └── placeholder.ts # SVG paper-textured frames └── assets/ ├── *.png # 16 collage plates (OpenDesign instance) ├── image-manifest.json # slot → file/dimensions/prompt mapping └── imagegen-prompts.md # human-readable prompt pack
design-systems/atelier-zero/DESIGN.md; extend the design system before adding a new ramp here.
data-reveal attributes from generated markup.Without them the page goes static and feels dead.
own stylesheet ordering — Atelier Zero relies on stylesheet-order cascade for paper texture and z-index of side rails.
fork; copy styles.css verbatim into app/globals.css so visual parity stays one-to-one.
design-systems/atelier-zero/DESIGN.md — token spec.apps/landing-page/ — deployable Astro static counterpart.design-templates/open-design-landing-deck/ — sibling slides skill that reuses this design system.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | pass→pass | 13,049 | 8,983 | -31% | 1 | 1 | 0% | 2,229 | 3,929 | +76% | 0 | 0 | — |
case-01 | fail→fail | 38,411 | 70,053 | +82% | 1 | 1 | 0% | 5,685 | 8,596 | +51% | 0 | 0 | — |
case-03 | pass→pass | 17,093 | 25,230 | +48% | 1 | 1 | 0% | 3,285 | 6,315 | +92% | 0 | 0 | — |
case-04 | fail→fail | 39,926 | 7,289 | -82% | 1 | 1 | 0% | 8,250 | 3,650 | -56% | 0 | 0 | — |
case-05 | fail→fail | 39,847 | 7,173 | -82% | 1 | 1 | 0% | 8,250 | 3,330 | -60% | 0 | 0 | — |
case-06 | fail→fail | 44,471 | 4,995 | -89% | 1 | 1 | 0% | 8,244 | 3,072 | -63% | 0 | 0 | — |
case-07 | fail→fail | 12,387 | 10,067 | -19% | 1 | 1 | 0% | 2,084 | 3,864 | +85% | 0 | 0 | — |
case-08 | fail→pass | 10,730 | 1,944 | -82% | 1 | 1 | 0% | 1,779 | 2,638 | +48% | 0 | 0 | — |
case-09 | fail→pass | 12,547 | 5,255 | -58% | 1 | 1 | 0% | 1,666 | 3,345 | +101% | 0 | 0 | — |
case-10 | fail→pass | 9,596 | 5,936 | -38% | 1 | 1 | 0% | 1,149 | 3,123 | +172% | 0 | 0 | — |
case-11 | fail→pass | 8,873 | 4,372 | -51% | 1 | 1 | 0% | 1,387 | 3,019 | +118% | 0 | 0 | — |
case-12 | pass→pass | 10,635 | 7,345 | -31% | 1 | 1 | 0% | 1,984 | 3,673 | +85% | 0 | 0 | — |
case-13 | fail→pass | 11,683 | 3,482 | -70% | 1 | 1 | 0% | 1,785 | 2,765 | +55% | 0 | 0 | — |
case-14 | fail→pass | 11,994 | 3,340 | -72% | 1 | 1 | 0% | 1,665 | 2,766 | +66% | 0 | 0 | — |
case-15 | fail→pass | 17,660 | 3,202 | -82% | 1 | 1 | 0% | 2,340 | 2,810 | +20% | 0 | 0 | — |
case-16 | fail→pass | 14,342 | 4,056 | -72% | 1 | 1 | 0% | 1,786 | 2,905 | +63% | 0 | 0 | — |
case-17 | fail→pass | 14,440 | 3,176 | -78% | 1 | 1 | 0% | 2,480 | 2,753 | +11% | 0 | 0 | — |
case-18 | fail→pass | 10,463 | 3,230 | -69% | 1 | 1 | 0% | 1,675 | 2,791 | +67% | 0 | 0 | — |
case-19 | fail→fail | 10,122 | 3,524 | -65% | 1 | 1 | 0% | 1,328 | 2,789 | +110% | 0 | 0 | — |
case-20 | fail→pass | 11,253 | 3,128 | -72% | 1 | 1 | 0% | 1,735 | 2,811 | +62% | 0 | 0 | — |
case-21 | pass→pass | 14,146 | 7,205 | -49% | 1 | 1 | 0% | 2,322 | 3,316 | +43% | 0 | 0 | — |
case-22 | fail→pass | 10,555 | 2,952 | -72% | 1 | 1 | 0% | 1,489 | 2,856 | +92% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +55 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.