Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build a complete typographic system -- modular or fluid type scale, line-height and tracking per size, weight roles, and vertical rhythm -- exported as tokens, Tailwind config, and CSS. Pairs with font-pairing-suggester (which picks the fonts; this builds the system they live in).
.claude/skills/onewave-ai-typography-scale-builder/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-22 | ✓→✓ | = Same ✓ | 26% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 46% | 0% |
Typography is the design system's skeleton: most products that look "off" have inconsistent type long before they have bad colors. Build the scale as a system with roles and math, then make the codebase actually use it. Input: the chosen fonts (or run font-pairing-suggester first), the product type (marketing site, dense app UI, long-form editorial), and the target range of devices.
clamp() per step so headings scale between viewport bounds without breakpoint jumps; fixed steps are fine for app UI.display, h1-h4, body-lg, body, body-sm, caption, overline, code. Each role = size step + line-height + tracking + weight + family, as one token. Components consume roles; nobody composes raw steps.text-wrap: balance on headings, and measure limits (65-75ch for body prose) so line length stays readable at every viewport.typography-tokens.json, Tailwind fontSize config (size + line-height + tracking tuples per role; v4 @theme), typography.css with role classes, and TYPOGRAPHY.md. Then map the step-1 inventory onto roles and replace -- with a per-case flag on anything that does not cleanly map (it is either a missing role or drift to delete).text-h2, never text-[28px] -- and no role means the component asks the system for one, not the developer for a number.clamp() minimum must stay readable on a 320px phone and the max must not blow the layout on ultrawide.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-22 | pass→pass | 16,336 | 14,787 | -9% | 1 | 1 | 0% | 2,833 | 3,580 | +26% | 0 | 0 | — |
case-01 | fail→fail | 38,557 | 37,498 | -3% | 1 | 1 | 0% | 8,247 | 9,113 | +11% | 0 | 0 | — |
case-02 | fail→fail | 24,779 | 31,370 | +27% | 1 | 1 | 0% | 5,269 | 8,245 | +56% | 0 | 0 | — |
case-03 | fail→fail | 29,492 | 33,495 | +14% | 1 | 1 | 0% | 5,967 | 8,793 | +47% | 0 | 0 | — |
case-04 | pass→pass | 20,378 | 24,219 | +19% | 1 | 1 | 0% | 3,799 | 5,557 | +46% | 0 | 0 | — |
case-05 | fail→fail | 19,879 | 23,109 | +16% | 1 | 1 | 0% | 3,798 | 5,493 | +45% | 0 | 0 | — |
case-06 | pass→pass | 12,451 | 12,345 | -1% | 1 | 1 | 0% | 2,130 | 3,099 | +45% | 0 | 0 | — |
case-07 | pass→pass | 13,064 | 14,047 | +8% | 1 | 1 | 0% | 2,270 | 3,540 | +56% | 0 | 0 | — |
case-08 | fail→pass | 13,874 | 12,882 | -7% | 1 | 1 | 0% | 2,518 | 3,283 | +30% | 0 | 0 | — |
case-17 | pass→pass | 16,606 | 12,301 | -26% | 1 | 1 | 0% | 2,678 | 2,771 | +3% | 0 | 0 | — |
case-09 | fail→pass | 14,910 | 14,137 | -5% | 1 | 1 | 0% | 2,518 | 3,451 | +37% | 0 | 0 | — |
case-10 | pass→pass | 14,651 | 14,463 | -1% | 1 | 1 | 0% | 2,353 | 3,171 | +35% | 0 | 0 | — |
case-11 | pass→pass | 13,289 | 10,843 | -18% | 1 | 1 | 0% | 2,175 | 2,467 | +13% | 0 | 0 | — |
case-12 | pass→pass | 18,779 | 18,029 | -4% | 1 | 1 | 0% | 3,174 | 3,830 | +21% | 0 | 0 | — |
case-13 | pass→pass | 13,915 | 9,876 | -29% | 1 | 1 | 0% | 2,190 | 2,509 | +15% | 0 | 0 | — |
case-14 | pass→pass | 16,096 | 19,264 | +20% | 1 | 1 | 0% | 3,017 | 4,891 | +62% | 0 | 0 | — |
case-15 | pass→pass | 7,273 | 8,332 | +15% | 1 | 1 | 0% | 1,326 | 2,381 | +80% | 0 | 0 | — |
case-16 | pass→pass | 14,032 | 15,263 | +9% | 1 | 1 | 0% | 2,420 | 3,740 | +55% | 0 | 0 | — |
case-18 | fail→pass | 13,783 | 17,485 | +27% | 1 | 1 | 0% | 2,412 | 3,942 | +63% | 0 | 0 | — |
case-19 | pass→pass | 15,317 | 13,186 | -14% | 1 | 1 | 0% | 2,431 | 3,066 | +26% | 0 | 0 | — |
case-20 | pass→pass | 16,295 | 16,960 | +4% | 1 | 1 | 0% | 3,259 | 4,331 | +33% | 0 | 0 | — |
case-21 | pass→pass | 12,430 | 11,595 | -7% | 1 | 1 | 0% | 2,516 | 3,043 | +21% | 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 +14 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.