Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create readable, performance-safe Three.js game visual effects. Use for attacks, impacts, damage feedback, status effects, spell trails, particles, shaders, telegraphs, quality tiers, and reduced-motion alternatives.
.claude/skills/mengto-create-game-vfx/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 2% | 0% |
Make the gameplay meaning visible before adding spectacle.
For every effect define trigger, owner, duration, gameplay meaning, camera-distance silhouette, color hierarchy, spawn cap, cleanup rule, and reduced-motion equivalent. Separate telegraph, contact, success, failure, and lingering status visuals.
Pool short-lived objects, reuse materials/geometry, cap particles, and avoid per-frame allocation. Use additive/transparency sparingly around important UI and targets. Make cleanup idempotent so reset, death, and pause cannot leak effects.
Test overlaps, multiple targets, rapid repetition, pause/resume, lowest quality, reduced motion, and touch viewports. Sample the live encounter for warnings, draw calls, frame time, and visual readability.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | pass→pass | 14,812 | 14,118 | -5% | 1 | 1 | 0% | 2,718 | 2,766 | +2% | 0 | 0 | — |
case-06 | pass→pass | 15,880 | 15,792 | -1% | 1 | 1 | 0% | 2,509 | 2,739 | +9% | 0 | 0 | — |
case-11 | pass→pass | 16,910 | 15,552 | -8% | 1 | 1 | 0% | 2,803 | 2,715 | -3% | 0 | 0 | — |
case-01 | fail→pass | 28,684 | 29,614 | +3% | 1 | 1 | 0% | 4,630 | 5,773 | +25% | 0 | 0 | — |
case-02 | fail→pass | 31,877 | 24,801 | -22% | 1 | 1 | 0% | 4,951 | 4,406 | -11% | 0 | 0 | — |
case-03 | fail→fail | 26,842 | 19,884 | -26% | 1 | 1 | 0% | 4,667 | 3,605 | -23% | 0 | 0 | — |
case-04 | fail→pass | 20,516 | 24,628 | +20% | 1 | 1 | 0% | 3,426 | 4,329 | +26% | 0 | 0 | — |
case-07 | fail→fail | 15,469 | 15,600 | +1% | 1 | 1 | 0% | 2,489 | 2,997 | +20% | 0 | 0 | — |
case-08 | pass→pass | 14,985 | 15,560 | +4% | 1 | 1 | 0% | 2,527 | 2,971 | +18% | 0 | 0 | — |
case-09 | pass→pass | 14,919 | 12,161 | -18% | 1 | 1 | 0% | 2,503 | 2,257 | -10% | 0 | 0 | — |
case-10 | pass→pass | 20,832 | 26,144 | +25% | 1 | 1 | 0% | 3,414 | 4,984 | +46% | 0 | 0 | — |
case-12 | pass→pass | 14,180 | 14,774 | +4% | 1 | 1 | 0% | 2,456 | 2,684 | +9% | 0 | 0 | — |
case-13 | pass→pass | 16,834 | 14,628 | -13% | 1 | 1 | 0% | 2,764 | 2,440 | -12% | 0 | 0 | — |
case-14 | fail→pass | 14,368 | 17,785 | +24% | 1 | 1 | 0% | 2,379 | 2,978 | +25% | 0 | 0 | — |
case-15 | fail→fail | 21,168 | 15,262 | -28% | 1 | 1 | 0% | 3,572 | 2,501 | -30% | 0 | 0 | — |
case-16 | pass→pass | 8,312 | 4,863 | -41% | 1 | 1 | 0% | 1,337 | 969 | -28% | 0 | 0 | — |
case-17 | pass→pass | 14,335 | 15,911 | +11% | 1 | 1 | 0% | 2,444 | 2,626 | +7% | 0 | 0 | — |
case-18 | pass→pass | 12,931 | 15,566 | +20% | 1 | 1 | 0% | 2,054 | 2,710 | +32% | 0 | 0 | — |
case-19 | pass→pass | 11,052 | 5,995 | -46% | 1 | 1 | 0% | 1,966 | 1,228 | -38% | 0 | 0 | — |
case-20 | pass→pass | 14,316 | 14,713 | +3% | 1 | 1 | 0% | 2,813 | 3,087 | +10% | 0 | 0 | — |
case-21 | pass→pass | 18,926 | 14,765 | -22% | 1 | 1 | 0% | 4,046 | 3,262 | -19% | 0 | 0 | — |
case-22 | pass→pass | 10,872 | 11,990 | +10% | 1 | 1 | 0% | 1,966 | 2,222 | +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 +18 percentage points is the difference between those two pass rates over the 22 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.