Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Rules for generating visual interfaces and parsing DESIGN.md files.
.claude/skills/bilal140202-design-system/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-11 | ✓→✓ | ＝ Same ✓ | 47% | 0% |
| case-20 | ✓→✓ | ＝ Same ✓ | -2% | 0% |
| case-21 | ✓→✓ | ＝ Same ✓ | 112% | 0% |
When generating graphical components, interfaces, or any front-end elements, you must follow this exact procedure:
Autonomously look for a DESIGN.md file in the root of the active workspace. This file dictates the absolute truth for typography, color tokens, margins, and the overall design system.
If the user specifies a known brand style (e.g., "make it look like Vercel" or "Vercel style") and no local DESIGN.md exists, fetch the standard from https://raw.githubusercontent.com/VoltAgent/awesome-design-md/main/design-md/[brand]/DESIGN.md using web search or URL reading tools.
If no DESIGN.md exists and no brand is specified:
bg-white/5 or bg-black/40 with backdrop-blur-xl.scale-105 duration-300).| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 24,699 | 24,790 | +0% | 1 | 1 | 0% | 6,163 | 6,470 | +5% | 0 | 0 | — |
case-13 | fail→fail | 18,332 | 26,974 | +47% | 1 | 1 | 0% | 4,114 | 6,463 | +57% | 0 | 0 | — |
case-01 | fail→fail | 18,121 | 15,027 | -17% | 1 | 1 | 0% | 4,442 | 3,905 | -12% | 0 | 0 | — |
case-02 | fail→fail | 18,070 | 19,642 | +9% | 1 | 1 | 0% | 3,906 | 4,954 | +27% | 0 | 0 | — |
case-04 | fail→pass | 22,360 | 27,109 | +21% | 1 | 1 | 0% | 5,033 | 6,334 | +26% | 0 | 0 | — |
case-05 | fail→fail | 23,207 | 23,783 | +2% | 1 | 1 | 0% | 5,187 | 5,805 | +12% | 0 | 0 | — |
case-06 | fail→fail | 21,631 | 26,450 | +22% | 1 | 1 | 0% | 4,460 | 6,460 | +45% | 0 | 0 | — |
case-07 | fail→fail | 19,048 | 17,533 | -8% | 1 | 1 | 0% | 4,211 | 4,309 | +2% | 0 | 0 | — |
case-08 | fail→fail | 7,214 | 13,393 | +86% | 1 | 1 | 0% | 1,658 | 3,259 | +97% | 0 | 0 | — |
case-14 | fail→fail | 17,716 | 20,808 | +17% | 1 | 1 | 0% | 4,044 | 5,226 | +29% | 0 | 0 | — |
case-09 | fail→fail | 11,505 | 13,442 | +17% | 1 | 1 | 0% | 2,651 | 3,266 | +23% | 0 | 0 | — |
case-10 | fail→fail | 13,160 | 17,220 | +31% | 1 | 1 | 0% | 2,879 | 4,059 | +41% | 0 | 0 | — |
case-11 | pass→pass | 16,276 | 21,335 | +31% | 1 | 1 | 0% | 3,820 | 5,622 | +47% | 0 | 0 | — |
case-12 | fail→fail | 23,099 | 24,549 | +6% | 1 | 1 | 0% | 5,614 | 6,462 | +15% | 0 | 0 | — |
case-15 | fail→fail | 23,521 | 24,740 | +5% | 1 | 1 | 0% | 6,153 | 6,460 | +5% | 0 | 0 | — |
case-16 | fail→pass | 12,092 | 12,704 | +5% | 1 | 1 | 0% | 2,621 | 2,968 | +13% | 0 | 0 | — |
case-17 | fail→fail | 18,755 | 23,200 | +24% | 1 | 1 | 0% | 3,696 | 4,868 | +32% | 0 | 0 | — |
case-18 | fail→fail | 14,577 | 18,780 | +29% | 1 | 1 | 0% | 3,140 | 4,562 | +45% | 0 | 0 | — |
case-19 | fail→fail | 15,793 | 24,916 | +58% | 1 | 1 | 0% | 3,489 | 6,052 | +73% | 0 | 0 | — |
case-20 | pass→pass | 15,358 | 15,451 | +1% | 1 | 1 | 0% | 3,568 | 3,486 | -2% | 0 | 0 | — |
case-21 | pass→pass | 5,157 | 10,653 | +107% | 1 | 1 | 0% | 1,114 | 2,363 | +112% | 0 | 0 | — |
case-22 | pass→pass | 10,051 | 9,682 | -4% | 1 | 1 | 0% | 2,053 | 2,456 | +20% | 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 +9 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.