Install any skill in seconds. Free to start, no credit card required.
Get Started Free →WebGL 风流体置换背景 + 顶部叠加金句, 适合视频片头 / landing hero / 海报
.claude/skills/nexu-io-frame-liquid-bg-hero/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 51% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 145% | 0% |
| case-22 | ✗→✓ | ▲ Improved | 13% | 0% |
【模板: 流体背景 Hero】 【意图】可作为视频片头帧、SaaS landing 顶部 hero、海报底图。WebGL 流体感, 但用 CSS / canvas 退化绘制, 确保单文件可双击打开。Inspired by hyperframes vfx-liquid-background。
【画布】1920×1080 (横) 或 1080×1920 (竖), 二选一。背景占满。
【流体背景 — 3 种实现, 按用户偏好选】
radial-gradient(...), 颜色取自调色板。@keyframes 平移 + scale + hue-rotate, 周期 8-14s, 错峰; 整个画面叠 mix-blend-mode: screen 或 overlay。backdrop-filter: blur(80px) 让边缘更糊。requestAnimationFrame 画 metaballs 或 simplex noise field。prefers-reduced-motion 时降回静态截图。regl 或 inline plain WebGL。u_time。【顶层文字层】
Source Serif Pro / Inter Tight / Manrope Black。#fafaf8 或 ink, 取决于背景明暗; 加 mix-blend-mode: difference 让它在任何流体颜色上都可读。【调色 — 4 选 1, 不要彩虹】
#ffb18a + #f78b4c + #d97757, 暖橙桃。#5ac8fa + #0a84ff + #1e3a8a, 海蓝。#a78bfa + #7c5cff + #1e1b4b, 极光紫。#86efac + #34d399 + #065f46, 苔森林。【设计细节】
Noto Serif SC (display) / Noto Sans SC (副标)。prefers-reduced-motion 关动效。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 29,086 | 28,132 | -3% | 1 | 1 | 0% | 6,212 | 7,000 | +13% | 0 | 0 | — |
case-02 | fail→fail | 26,644 | 28,359 | +6% | 1 | 1 | 0% | 6,221 | 7,009 | +13% | 0 | 0 | — |
case-03 | fail→fail | 27,977 | 26,422 | -6% | 1 | 1 | 0% | 6,226 | 7,014 | +13% | 0 | 0 | — |
case-04 | fail→fail | 28,403 | 26,573 | -6% | 1 | 1 | 0% | 6,210 | 6,998 | +13% | 0 | 0 | — |
case-05 | fail→pass | 22,387 | 26,519 | +18% | 1 | 1 | 0% | 4,627 | 6,985 | +51% | 0 | 0 | — |
case-06 | pass→fail | 22,590 | 27,973 | +24% | 1 | 1 | 0% | 4,964 | 6,963 | +40% | 0 | 0 | — |
case-07 | pass→pass | 18,898 | 26,718 | +41% | 1 | 1 | 0% | 4,738 | 6,973 | +47% | 0 | 0 | — |
case-08 | fail→pass | 28,196 | 27,715 | -2% | 1 | 1 | 0% | 6,199 | 6,987 | +13% | 0 | 0 | — |
case-09 | fail→pass | 19,094 | 28,995 | +52% | 1 | 1 | 0% | 4,162 | 6,972 | +68% | 0 | 0 | — |
case-10 | pass→pass | 17,479 | 26,801 | +53% | 1 | 1 | 0% | 3,940 | 6,968 | +77% | 0 | 0 | — |
case-11 | fail→fail | 19,369 | 26,547 | +37% | 1 | 1 | 0% | 4,282 | 6,973 | +63% | 0 | 0 | — |
case-12 | pass→pass | 26,463 | 27,148 | +3% | 1 | 1 | 0% | 6,182 | 6,970 | +13% | 0 | 0 | — |
case-13 | pass→pass | 23,942 | 25,829 | +8% | 1 | 1 | 0% | 5,294 | 6,971 | +32% | 0 | 0 | — |
case-14 | fail→fail | 31,517 | 27,102 | -14% | 1 | 1 | 0% | 6,172 | 6,960 | +13% | 0 | 0 | — |
case-15 | pass→pass | 9,625 | 28,104 | +192% | 1 | 1 | 0% | 1,679 | 6,993 | +316% | 0 | 0 | — |
case-16 | pass→pass | 30,502 | 29,683 | -3% | 1 | 1 | 0% | 6,183 | 6,971 | +13% | 0 | 0 | — |
case-17 | fail→pass | 14,500 | 28,033 | +93% | 1 | 1 | 0% | 2,842 | 6,964 | +145% | 0 | 0 | — |
case-18 | pass→pass | 27,024 | 29,057 | +8% | 1 | 1 | 0% | 6,185 | 6,973 | +13% | 0 | 0 | — |
case-19 | pass→pass | 21,962 | 29,425 | +34% | 1 | 1 | 0% | 5,051 | 6,966 | +38% | 0 | 0 | — |
case-20 | pass→fail | 25,258 | 24,571 | -3% | 1 | 1 | 0% | 6,179 | 6,967 | +13% | 0 | 0 | — |
case-21 | pass→pass | 8,959 | 26,450 | +195% | 1 | 1 | 0% | 2,286 | 6,955 | +204% | 0 | 0 | — |
case-22 | fail→pass | 28,989 | 30,676 | +6% | 1 | 1 | 0% | 6,168 | 6,956 | +13% | 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. 2 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.