Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use this skill when generating Open Design artifacts that should follow the HiCatcat G2 AR glasses HUD design system. This plugin packages a browser-reviewable UI kit, local token CSS, and modular React role components.
.claude/skills/nexu-io-g2-design-system-ui-kit/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 140% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -29% | 0% |
Use this skill when generating OpenDesign artifacts that should follow the HiCatcat G2 AR glasses HUD design system. This plugin packages a browser-reviewable UI kit, local token CSS, and modular React role components.
index.html - Browser-reviewable entry that loads ./colors_and_type.css, React, ReactDOM, Babel, and the component files.colors_and_type.css - Source-backed G2 token CSS for dark HUD color, type, spacing, radius, icon, stroke, and compact/regular display modes.components/App.jsx - App shell that composes the G2 HUD surface.components/Sidebar.jsx - Compact system rail for mode, call, messages, AI, and device controls.components/AssistantsList.jsx - AI card and quick status pattern.components/ChatArea.jsx - Primary message, call, and teleprompter workspace.components/InputBar.jsx - Bottom command-entry surface.components/MessageBubble.jsx - Message, note, and assistant response unit.README.md - Human-readable package guide and validation notes.Open index.html directly for visual review. Copy component files into a React prototype when building product-like artifacts. Keep colors_and_type.css loaded before the components so color, type, spacing, radius, stroke, and mode variables resolve through the G2 token contract.
Use G2 as a dark AR glasses control interface, not a generic mobile app, SaaS dashboard, or marketing page. Preserve these source-backed cues:
#333333 controls, borders, and disabled surfaces.#0D76FF selected state.#0D76FFA6 focus and touch feedback.data-g2-mode="regular" for 640 mode and data-g2-mode="compact" for 320 mode.Do not invent off-token accent colors, glassmorphism, or landing-page hero compositions.
Formalized by OpenDesign from candidate 33633026-3f08-4a00-86c8-1bbd1d3e1ce4 and corrected so every referenced file is packaged in this plugin folder.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→fail | 27,913 | 35,271 | +26% | 1 | 1 | 0% | 4,189 | 8,782 | +110% | 0 | 0 | — |
case-01 | fail→fail | 82,351 | 50,928 | -38% | 1 | 1 | 0% | 8,235 | 8,800 | +7% | 0 | 0 | — |
case-02 | fail→pass | 35,661 | 51,283 | +44% | 1 | 1 | 0% | 8,227 | 8,792 | +7% | 0 | 0 | — |
case-03 | fail→fail | 50,866 | 36,527 | -28% | 1 | 1 | 0% | 8,218 | 8,783 | +7% | 0 | 0 | — |
case-04 | fail→pass | 48,579 | 40,833 | -16% | 1 | 1 | 0% | 3,661 | 8,801 | +140% | 0 | 0 | — |
case-05 | fail→pass | 26,763 | 54,899 | +105% | 1 | 1 | 0% | 4,169 | 5,655 | +36% | 0 | 0 | — |
case-06 | fail→fail | 21,826 | 24,609 | +13% | 1 | 1 | 0% | 3,272 | 5,033 | +54% | 0 | 0 | — |
case-07 | fail→fail | 48,568 | 49,683 | +2% | 1 | 1 | 0% | 7,787 | 8,788 | +13% | 0 | 0 | — |
case-08 | fail→fail | 49,702 | 52,732 | +6% | 1 | 1 | 0% | 8,218 | 8,783 | +7% | 0 | 0 | — |
case-09 | fail→fail | 38,938 | 46,713 | +20% | 1 | 1 | 0% | 8,215 | 8,780 | +7% | 0 | 0 | — |
case-11 | fail→fail | 26,361 | 40,614 | +54% | 1 | 1 | 0% | 5,018 | 8,779 | +75% | 0 | 0 | — |
case-12 | fail→fail | 37,685 | 35,923 | -5% | 1 | 1 | 0% | 8,208 | 8,773 | +7% | 0 | 0 | — |
case-13 | fail→pass | 27,379 | 23,933 | -13% | 1 | 1 | 0% | 3,798 | 3,448 | -9% | 0 | 0 | — |
case-14 | fail→pass | 16,244 | 6,067 | -63% | 1 | 1 | 0% | 1,982 | 1,398 | -29% | 0 | 0 | — |
case-15 | fail→fail | 34,659 | 36,505 | +5% | 1 | 1 | 0% | 7,483 | 8,735 | +17% | 0 | 0 | — |
case-16 | fail→pass | 19,146 | 13,716 | -28% | 1 | 1 | 0% | 2,484 | 2,980 | +20% | 0 | 0 | — |
case-17 | fail→pass | 19,073 | 44,100 | +131% | 1 | 1 | 0% | 2,732 | 7,721 | +183% | 0 | 0 | — |
case-18 | fail→pass | 17,837 | 64,547 | +262% | 1 | 1 | 0% | 3,263 | 8,782 | +169% | 0 | 0 | — |
case-19 | fail→pass | 12,087 | 7,607 | -37% | 1 | 1 | 0% | 2,204 | 1,863 | -15% | 0 | 0 | — |
case-20 | pass→pass | 36,756 | 58,004 | +58% | 1 | 1 | 0% | 8,222 | 8,787 | +7% | 0 | 0 | — |
case-21 | pass→pass | 29,575 | 50,766 | +72% | 1 | 1 | 0% | 5,861 | 8,782 | +50% | 0 | 0 | — |
case-22 | pass→pass | 30,888 | 42,886 | +39% | 1 | 1 | 0% | 4,209 | 8,788 | +109% | 0 | 0 | — |
case-23 | fail→pass | 15,137 | 20,289 | +34% | 1 | 1 | 0% | 1,972 | 1,026 | -48% | 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. The headline lift of +43 percentage points is the difference between those two pass rates over the 23 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.