Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Extract design tokens (color / typography / spacing) from imported source code, screenshots, or Figma exports into the canonical token bag token-map consumes.
.claude/skills/nexu-io-design-extract/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 2134% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 883% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -57% | 0% |
Spec §10 / §21.3.2: tokens scattered across a repo's CSS / theme files / Tailwind config / styled-components helpers / SCSS partials need to be lifted into one bag before token-map can crosswalk them onto the active OD design system. This atom does the lifting; the input shape is intentionally generic so the same atom serves both code-migration and figma-migration when fed different sources.
code/index.json from code-import (code-migration), ORfigma/tree.json from figma-extract (figma-migration), ORtextproject-cwd/ └── code/ └── tokens.json # { colors[], typography[], spacing[], radius[], shadow[] }
Each token entry carries:
jsonc{ "kind": "color", "name": "primary-500", // optional source name "value": "#5b8def", // canonical value "sources": ["styles/global.css:42", "tailwind.config.js:24"], "usage": ["Header.tsx", "Footer.tsx"] }
sources[] and usage[] are the audit trail token-map.unmatched.json references when a target token can't be found.
The atom completes when code/tokens.json exists. Empty token bags emit a warning event but do not abort — token-map handles the empty case by skipping its mapping pass.
color: '#fff') aretokens for this atom's purposes; record them with kind:'color' and a synthetic name so they don't disappear into the noise.
active theme — never list bg-blue-500 and bg-blue-600 as one token.
Implemented by the daemon runner in apps/daemon/src/plugins/atoms/design-extract.ts. It scans the indexed source files, extracts CSS, Tailwind, and JavaScript token evidence, and writes code/tokens.json.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,304 | 38,017 | +617% | 1 | 1 | 0% | 229 | 5,117 | +2134% | 0 | 0 | — |
case-02 | fail→pass | 7,156 | 14,987 | +109% | 1 | 1 | 0% | 305 | 2,998 | +883% | 0 | 0 | — |
case-03 | pass→pass | 37,468 | 16,692 | -55% | 1 | 1 | 0% | 5,960 | 3,926 | -34% | 0 | 0 | — |
case-04 | fail→fail | 23,118 | 20,443 | -12% | 1 | 1 | 0% | 4,835 | 3,918 | -19% | 0 | 0 | — |
case-05 | fail→fail | 12,908 | 16,064 | +24% | 1 | 1 | 0% | 2,011 | 3,054 | +52% | 0 | 0 | — |
case-06 | fail→fail | 32,969 | 26,112 | -21% | 1 | 1 | 0% | 3,774 | 4,635 | +23% | 0 | 0 | — |
case-07 | fail→pass | 14,325 | 7,625 | -47% | 1 | 1 | 0% | 1,893 | 1,620 | -14% | 0 | 0 | — |
case-08 | pass→pass | 12,366 | 4,663 | -62% | 1 | 1 | 0% | 1,739 | 1,429 | -18% | 0 | 0 | — |
case-09 | pass→pass | 13,256 | 2,872 | -78% | 1 | 1 | 0% | 1,790 | 973 | -46% | 0 | 0 | — |
case-10 | fail→pass | 12,868 | 10,866 | -16% | 1 | 1 | 0% | 1,866 | 1,993 | +7% | 0 | 0 | — |
case-11 | fail→pass | 12,341 | 2,606 | -79% | 1 | 1 | 0% | 2,162 | 924 | -57% | 0 | 0 | — |
case-12 | fail→pass | 11,988 | 4,136 | -65% | 1 | 1 | 0% | 1,599 | 1,147 | -28% | 0 | 0 | — |
case-13 | fail→pass | 5,005 | 2,566 | -49% | 1 | 1 | 0% | 941 | 1,026 | +9% | 0 | 0 | — |
case-14 | pass→pass | 12,369 | 4,520 | -63% | 1 | 1 | 0% | 2,002 | 1,191 | -41% | 0 | 0 | — |
case-15 | pass→pass | 16,504 | 5,854 | -65% | 1 | 1 | 0% | 2,008 | 1,406 | -30% | 0 | 0 | — |
case-16 | fail→pass | 12,970 | 9,309 | -28% | 1 | 1 | 0% | 1,726 | 1,804 | +5% | 0 | 0 | — |
case-17 | pass→pass | 8,595 | 3,217 | -63% | 1 | 1 | 0% | 1,369 | 1,091 | -20% | 0 | 0 | — |
case-18 | pass→pass | 14,109 | 6,540 | -54% | 1 | 1 | 0% | 1,962 | 1,466 | -25% | 0 | 0 | — |
case-19 | fail→pass | 7,318 | 2,323 | -68% | 1 | 1 | 0% | 1,238 | 770 | -38% | 0 | 0 | — |
case-20 | fail→pass | 18,143 | 4,146 | -77% | 1 | 1 | 0% | 2,691 | 1,209 | -55% | 0 | 0 | — |
case-21 | pass→pass | 14,072 | 3,851 | -73% | 1 | 1 | 0% | 2,161 | 1,241 | -43% | 0 | 0 | — |
case-22 | fail→pass | 23,915 | 2,546 | -89% | 1 | 1 | 0% | 2,048 | 812 | -60% | 0 | 0 | — |
case-23 | fail→pass | 8,319 | 2,869 | -66% | 1 | 1 | 0% | 1,376 | 930 | -32% | 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. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +52 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.