Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →拟真 macOS 通知 banner + app icon + 标题正文, 适合 video overlay / 产品发布预告
.claude/skills/nexu-io-frame-macos-notification/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 52% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 86% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 57% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -3% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 13% | 0% |
【模板: macOS 通知横幅】 【意图】把一段公告 / 消息 / 提示渲染成 macOS Big Sur+ 风格的通知横幅, 适合视频角落叠加、产品发布预告、社媒图。Inspired by hyperframes macos-notification。
【画布】两种用法:
【横幅结构】
background: rgba(245,245,247,0.78) + backdrop-filter: blur(40px) saturate(180%); 暗色版 rgba(28,28,30,0.78)。rgba(0,0,0,0.06) (light) / rgba(255,255,255,0.08) (dark); 顶部加 1px 亮 highlight rgba(255,255,255,0.5)。0 10px 40px rgba(0,0,0,0.18), 0 2px 6px rgba(0,0,0,0.08)。【内容】
now 或具体时间 (12px, opacity 0.6) — 两端对齐。【字体】
SF Pro Text → fallback Inter / system-ui; 中文用 PingFang SC / Noto Sans SC。【可选附加】
transform: translateX(110%)→0, 200ms ease-out; 可被 prefers-reduced-motion 关闭。【设计细节】
backdrop-filter Safari 需要 -webkit- 前缀。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 25,525 | 23,584 | -8% | 1 | 1 | 0% | 5,119 | 6,046 | +18% | 0 | 0 | — |
case-02 | fail→fail | 28,293 | 19,649 | -31% | 1 | 1 | 0% | 6,210 | 5,373 | -13% | 0 | 0 | — |
case-03 | fail→fail | 28,652 | 26,774 | -7% | 1 | 1 | 0% | 6,200 | 6,986 | +13% | 0 | 0 | — |
case-04 | pass→pass | 28,199 | 22,516 | -20% | 1 | 1 | 0% | 6,197 | 6,028 | -3% | 0 | 0 | — |
case-05 | pass→fail | 18,996 | 27,593 | +45% | 1 | 1 | 0% | 4,427 | 6,956 | +57% | 0 | 0 | — |
case-06 | pass→pass | 28,403 | 27,947 | -2% | 1 | 1 | 0% | 6,167 | 6,953 | +13% | 0 | 0 | — |
case-07 | fail→fail | 22,255 | 19,361 | -13% | 1 | 1 | 0% | 4,795 | 4,950 | +3% | 0 | 0 | — |
case-08 | fail→fail | 15,348 | 19,891 | +30% | 1 | 1 | 0% | 3,626 | 5,194 | +43% | 0 | 0 | — |
case-09 | fail→fail | 19,163 | 23,411 | +22% | 1 | 1 | 0% | 4,363 | 6,080 | +39% | 0 | 0 | — |
case-10 | fail→fail | 21,018 | 26,239 | +25% | 1 | 1 | 0% | 4,813 | 6,957 | +45% | 0 | 0 | — |
case-11 | fail→fail | 21,300 | 27,748 | +30% | 1 | 1 | 0% | 4,733 | 6,951 | +47% | 0 | 0 | — |
case-12 | fail→fail | 14,706 | 16,872 | +15% | 1 | 1 | 0% | 3,508 | 4,680 | +33% | 0 | 0 | — |
case-13 | fail→fail | 17,920 | 23,767 | +33% | 1 | 1 | 0% | 4,466 | 5,700 | +28% | 0 | 0 | — |
case-14 | fail→pass | 22,704 | 25,545 | +13% | 1 | 1 | 0% | 4,565 | 6,947 | +52% | 0 | 0 | — |
case-15 | fail→fail | 18,521 | 25,796 | +39% | 1 | 1 | 0% | 3,830 | 6,957 | +82% | 0 | 0 | — |
case-16 | pass→pass | 18,225 | 28,861 | +58% | 1 | 1 | 0% | 3,953 | 6,951 | +76% | 0 | 0 | — |
case-17 | pass→pass | 23,450 | 26,073 | +11% | 1 | 1 | 0% | 5,053 | 6,508 | +29% | 0 | 0 | — |
case-18 | fail→fail | 27,084 | 25,347 | -6% | 1 | 1 | 0% | 6,081 | 6,538 | +8% | 0 | 0 | — |
case-19 | fail→fail | 23,690 | 27,337 | +15% | 1 | 1 | 0% | 5,176 | 6,942 | +34% | 0 | 0 | — |
case-20 | fail→pass | 16,007 | 23,634 | +48% | 1 | 1 | 0% | 3,592 | 6,690 | +86% | 0 | 0 | — |
case-21 | fail→fail | 14,551 | 24,585 | +69% | 1 | 1 | 0% | 3,418 | 6,485 | +90% | 0 | 0 | — |
case-22 | pass→pass | 19,281 | 25,503 | +32% | 1 | 1 | 0% | 4,456 | 6,650 | +49% | 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 +5 percentage points is the difference between those two pass rates over the 22 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.