Install any skill in seconds. Free to start, no credit card required.
Get Started Free →This file is generated during setup and should be treated as source evidence for the design-system project. Use it before writing or revising DESIGN.md, previews, tokens, UI kit examples, or assets.
.claude/skills/nexu-io-design-system-source-context/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-20 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 119% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 62% | 0% |
This file is generated during setup and should be treated as source evidence for the design-system project. Use it before writing or revising DESIGN.md, previews, tokens, UI kit examples, or assets.
Canonical design-system title: Claude / Anthropic 风格设计系统
非官方 Claude / Anthropic 风格整理版:温暖、克制、编辑感、适合阅读与深度思考的 AI 工作台。当前 canonical 依据包括 assets/Claude-anthropic-DESIGN风格设计系统.md,以及 2026-06-29 抽查的 Anthropic 官网首页、Newsroom、Company 页面公开 CSS。颜色和交互以官网 token 命名校准:ivory-light #faf9f5、ivory-medium #f0eee6、ivory-dark #e8e6dc、slate-dark #141413、clay #d97757。
Connector status: GitHub connector is not configured; repository intake will use local git credentials or authenticated GitHub CLI when possible.
Linked folders readable by the local agent: none.
Copied browser-selected code snapshot files under context/local-code/: none.
Figma files selected: none.
Decoded Figma snapshots: none. Fonts, logos, and assets selected:
Uploaded brand asset files under assets/:
No additional notes provided.
/design-systems/create only collected setup inputs. All GitHub extraction, website/source URL review, local evidence intake, source reading, design-system construction, package audit, and artifact writes should happen inside this project workspace.https.## Preview Manifest listing every generated preview/*.html card, and reuse workflow, similar to Claude Design exports.name, description, and user-invocable, plus Claude-style reusable skill sections: What is inside, Source context, When to use this skill, How to use, and Design system highlights. The usage guidance should point agents at README.md, DESIGN.md, colors_and_type.css, preview/, assets/, build/, fonts/, source_examples/, and ui_kits/app/.ui_kits/app/ paths, not old scaffold names such as preview/typography-scale.html or ui_kits/generated_interface/.ui_kits/app/components/ when the source snapshots include representative app shells, navigation, chat/input surfaces, or reusable components. ui_kits/app/README.md should explain structure, component files, usage, design notes, and source basis. ui_kits/app/index.html must load ../../colors_and_type.css, must load/import/compose the modular component files, and must mount/render the composed interface instead of staying as a standalone generic static mock or disconnected script list. If the entry directly loads .jsx/.tsx files, include React, ReactDOM, and Babel standalone scripts and expose each loaded component as window.ComponentName / globalThis.ComponentName, or write compiled browser-ready JavaScript instead. For chat/workspace evidence, cover app shell, sidebar/navigation, assistant/list rail, chat area, input bar/composer, and message bubble/comment roles; the app shell component must compose those roles into one product-like surface. Placeholder component shells are not sufficient.Claude-style UI-kit entry contract:
ui_kits/app/components/*.jsx or *.tsx files exist, ui_kits/app/index.html must behave like a runnable browser entry, not a static mock.../../colors_and_type.css, create a #root, load each component script from components/, then render the composed App component.App.jsx must assign window.App = App (or globalThis.App = App), and every directly loaded component file must expose the same browser global for its component name.html<script src="https://unpkg.com/react@18.3.1/umd/react.development.js"></script> <script src="https://unpkg.com/react-dom@18.3.1/umd/react-dom.development.js"></script> <script src="https://unpkg.com/@babel/standalone@7.29.0/babel.min.js"></script> <link rel="stylesheet" href="../../colors_and_type.css"> <div id="root"></div> <script type="text/babel" src="components/Sidebar.jsx"></script> <script type="text/babel" src="components/AssistantsList.jsx"></script> <script type="text/babel" src="components/ChatArea.jsx"></script> <script type="text/babel" src="components/MessageBubble.jsx"></script> <script type="text/babel" src="components/InputBar.jsx"></script> <script type="text/babel" src="components/App.jsx"></script> <script type="text/babel"> const { App } = window; const root = ReactDOM.createRoot(document.getElementById('root')); root.render(<App />); </script>
Claude-style build asset contract:
context/.../files/build/..., create a root build/ directory and copy representative runtime assets there with their original filenames and path intent, such as build/icon.png, build/logo.png, build/tray_icon.png, and build/icon.ico.context/.../files/... snapshots. Do not redraw, re-encode, optimize, or substitute generated placeholders for files that the evidence already captured.assets/. assets/ may include convenience aliases, but root build/ must preserve the source runtime files for future agents and package consumers.preview/brand-assets.html should reference at least some real preserved files from build/ or assets/ with <img>, <picture>, <object>, or CSS url(...), and README.md / SKILL.md should mention build/ in the package manifest when it exists.github-design-context command, not direct connector tree/content/raw tool calls. The command tries this-device git first, authenticated GitHub CLI second, and connector-platform fallback only when local access cannot read the repository.local-design-context command, which writes a local evidence note and snapshots under context/local-code/ before final design-system rules are drafted."$OD_NODE_BIN" "$OD_BIN" tools connectors design-system-package-audit --path . --fail-on-warnings and fix every reported error or warning.Formalized by OpenDesign from candidate 242e9a72-9a69-4de5-91a4-67563dcf7db1.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-20 | fail→pass | 14,824 | 2,296 | -85% | 1 | 1 | 0% | 2,023 | 2,437 | +20% | 0 | 0 | — |
case-01 | fail→fail | 50,300 | 14,957 | -70% | 1 | 1 | 0% | 8,269 | 3,418 | -59% | 0 | 0 | — |
case-02 | fail→fail | 42,114 | 8,016 | -81% | 1 | 1 | 0% | 8,267 | 2,952 | -64% | 0 | 0 | — |
case-03 | fail→fail | 8,368 | 13,545 | +62% | 1 | 1 | 0% | 229 | 2,940 | +1184% | 0 | 0 | — |
case-04 | pass→fail | 16,255 | 22,040 | +36% | 1 | 1 | 0% | 3,498 | 5,677 | +62% | 0 | 0 | — |
case-05 | pass→pass | 10,445 | 13,189 | +26% | 1 | 1 | 0% | 1,811 | 4,516 | +149% | 0 | 0 | — |
case-06 | pass→pass | 11,851 | 9,311 | -21% | 1 | 1 | 0% | 2,164 | 3,834 | +77% | 0 | 0 | — |
case-07 | fail→pass | 14,847 | 5,540 | -63% | 1 | 1 | 0% | 1,869 | 3,021 | +62% | 0 | 0 | — |
case-08 | fail→pass | 9,109 | 6,027 | -34% | 1 | 1 | 0% | 1,090 | 2,388 | +119% | 0 | 0 | — |
case-09 | fail→pass | 16,223 | 9,666 | -40% | 1 | 1 | 0% | 2,303 | 3,125 | +36% | 0 | 0 | — |
case-10 | pass→fail | 12,718 | 6,189 | -51% | 1 | 1 | 0% | 1,907 | 2,580 | +35% | 0 | 0 | — |
case-11 | pass→pass | 9,502 | 4,427 | -53% | 1 | 1 | 0% | 1,308 | 2,993 | +129% | 0 | 0 | — |
case-12 | fail→pass | 12,496 | 7,337 | -41% | 1 | 1 | 0% | 2,264 | 3,669 | +62% | 0 | 0 | — |
case-13 | fail→fail | 16,853 | 7,405 | -56% | 1 | 1 | 0% | 2,407 | 3,563 | +48% | 0 | 0 | — |
case-14 | fail→pass | 22,761 | 3,697 | -84% | 1 | 1 | 0% | 2,215 | 2,481 | +12% | 0 | 0 | — |
case-15 | fail→pass | 18,049 | 9,556 | -47% | 1 | 1 | 0% | 2,813 | 3,350 | +19% | 0 | 0 | — |
case-16 | fail→fail | 6,189 | 2,331 | -62% | 1 | 1 | 0% | 1,060 | 2,486 | +135% | 0 | 0 | — |
case-17 | pass→pass | 13,297 | 8,596 | -35% | 1 | 1 | 0% | 2,035 | 2,976 | +46% | 0 | 0 | — |
case-18 | fail→pass | 10,696 | 1,888 | -82% | 1 | 1 | 0% | 1,625 | 2,318 | +43% | 0 | 0 | — |
case-19 | fail→fail | 18,004 | 4,920 | -73% | 1 | 1 | 0% | 2,315 | 2,818 | +22% | 0 | 0 | — |
case-21 | fail→pass | 5,135 | 2,231 | -57% | 1 | 1 | 0% | 731 | 2,517 | +244% | 0 | 0 | — |
case-22 | fail→pass | 4,765 | 2,658 | -44% | 1 | 1 | 0% | 615 | 2,548 | +314% | 0 | 0 | — |
case-23 | fail→pass | 15,594 | 2,328 | -85% | 1 | 1 | 0% | 2,069 | 2,578 | +25% | 0 | 0 | — |
case-24 | fail→pass | 12,376 | 4,163 | -66% | 1 | 1 | 0% | 1,964 | 2,728 | +39% | 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. 24 cases were attempted, and 20 counted toward the lift figure. The other 4 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 +42 percentage points is the difference between those two pass rates over the 20 comparable cases. 3 cases got worse with the skill loaded, and they are 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.