Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Profile, diagnose, and improve Three.js or WebGL game performance without regressing gameplay. Use for frame-time drops, CPU/GPU pressure, draw calls, texture and geometry budgets, animation loops, adaptive quality, mobile performance, and browser performance verification.
.claude/skills/mengto-optimize-threejs-games/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-14 | ✓→✓ | = Same ✓ | -12% | 0% |
| case-04 | ✓→✓ | = Same ✓ | -7% | 0% |
Measure before changing behavior, then validate the same gameplay path after every optimization.
Choose a deterministic representative encounter and record device, viewport, quality level, player position, enemies, active effects, frame-time sample, draw calls, triangles, texture count, and warnings. Compare like with like.
Reuse geometry/materials, pool transient effects, cull inactive/offscreen work, throttle noncritical UI updates, cap particle counts, avoid per-frame allocation, and update only changed transforms. Keep render quality settings explicit and reversible. Degrade decorative effects before combat readability or controls.
Re-run the original encounter and verify frame time, visual correctness, collision/contact behavior, memory stability, mobile controls, reduced motion, and console health. Close temporary servers, benchmarks, and browser tabs once they are no longer needed.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→fail | 15,766 | 9,394 | -40% | 1 | 1 | 0% | 2,617 | 1,961 | -25% | 0 | 0 | — |
case-14 | pass→pass | 13,933 | 10,445 | -25% | 1 | 1 | 0% | 2,517 | 2,213 | -12% | 0 | 0 | — |
case-01 | fail→fail | 26,301 | 20,496 | -22% | 1 | 1 | 0% | 5,251 | 4,337 | -17% | 0 | 0 | — |
case-02 | fail→fail | 21,006 | 16,832 | -20% | 1 | 1 | 0% | 4,107 | 3,167 | -23% | 0 | 0 | — |
case-03 | fail→fail | 20,910 | 17,352 | -17% | 1 | 1 | 0% | 3,866 | 3,279 | -15% | 0 | 0 | — |
case-04 | pass→pass | 18,058 | 16,297 | -10% | 1 | 1 | 0% | 4,090 | 3,791 | -7% | 0 | 0 | — |
case-05 | pass→pass | 14,848 | 11,736 | -21% | 1 | 1 | 0% | 3,569 | 2,905 | -19% | 0 | 0 | — |
case-06 | pass→pass | 19,145 | 18,220 | -5% | 1 | 1 | 0% | 3,353 | 3,483 | +4% | 0 | 0 | — |
case-08 | pass→pass | 15,878 | 12,853 | -19% | 1 | 1 | 0% | 2,655 | 2,458 | -7% | 0 | 0 | — |
case-09 | pass→pass | 13,308 | 13,473 | +1% | 1 | 1 | 0% | 2,335 | 2,615 | +12% | 0 | 0 | — |
case-10 | fail→fail | 16,077 | 13,031 | -19% | 1 | 1 | 0% | 2,781 | 2,646 | -5% | 0 | 0 | — |
case-11 | fail→fail | 11,912 | 8,019 | -33% | 1 | 1 | 0% | 2,035 | 1,620 | -20% | 0 | 0 | — |
case-12 | fail→pass | 14,662 | 8,131 | -45% | 1 | 1 | 0% | 2,535 | 1,651 | -35% | 0 | 0 | — |
case-13 | pass→pass | 12,810 | 9,742 | -24% | 1 | 1 | 0% | 2,419 | 2,188 | -10% | 0 | 0 | — |
case-15 | pass→pass | 8,880 | 3,460 | -61% | 1 | 1 | 0% | 1,495 | 885 | -41% | 0 | 0 | — |
case-16 | pass→pass | 10,732 | 7,399 | -31% | 1 | 1 | 0% | 1,975 | 1,569 | -21% | 0 | 0 | — |
case-17 | pass→pass | 11,074 | 9,623 | -13% | 1 | 1 | 0% | 1,952 | 1,833 | -6% | 0 | 0 | — |
case-18 | fail→pass | 10,566 | 7,202 | -32% | 1 | 1 | 0% | 1,867 | 1,418 | -24% | 0 | 0 | — |
case-19 | fail→pass | 13,478 | 10,554 | -22% | 1 | 1 | 0% | 2,333 | 2,081 | -11% | 0 | 0 | — |
case-20 | pass→pass | 9,409 | 6,661 | -29% | 1 | 1 | 0% | 1,575 | 1,336 | -15% | 0 | 0 | — |
case-21 | pass→pass | 12,793 | 9,346 | -27% | 1 | 1 | 0% | 2,277 | 1,891 | -17% | 0 | 0 | — |
case-22 | fail→fail | 11,144 | 6,387 | -43% | 1 | 1 | 0% | 1,887 | 1,265 | -33% | 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. 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.