Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build, tune, or test a Three.js game for mobile web. Use for touch movement, action controls, target selection, touch inventory, safe areas, portrait/landscape layouts, responsive HUD, battery/performance budgets, and real mobile browser QA.
.claude/skills/mengto-build-mobile-threejs-games/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-19 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 6% | 0% |
Treat mobile as a primary control and performance surface.
Provide a clear movement control, reachable primary actions, target selection, cancel/reset paths, and non-drag alternatives for inventory. Keep controls out of unsafe areas, avoid overlapping action zones, and show feedback for held, unavailable, and queued actions.
Design for representative portrait and landscape viewports. Keep vital status, target data, inventory, tooltips, and dialogs legible without covering the playfield. Support orientation changes without resetting gameplay state.
Use device-appropriate quality defaults and adaptive decorative effects. Test touch targets, multi-touch, long sessions, audio unlock, pause/background return, reduced motion, low-end quality, and actual mobile-size viewports in the repository-approved browser.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-21 | pass→fail | 15,771 | 15,015 | -5% | 1 | 1 | 0% | 2,592 | 2,895 | +12% | 0 | 0 | — |
case-22 | pass→pass | 14,436 | 15,580 | +8% | 1 | 1 | 0% | 2,501 | 2,910 | +16% | 0 | 0 | — |
case-19 | fail→pass | 14,981 | 8,885 | -41% | 1 | 1 | 0% | 2,689 | 1,855 | -31% | 0 | 0 | — |
case-01 | fail→pass | 20,621 | 26,104 | +27% | 1 | 1 | 0% | 4,567 | 6,363 | +39% | 0 | 0 | — |
case-02 | fail→fail | 23,694 | 20,279 | -14% | 1 | 1 | 0% | 4,127 | 4,094 | -1% | 0 | 0 | — |
case-03 | fail→pass | 19,456 | 16,135 | -17% | 1 | 1 | 0% | 3,240 | 3,090 | -5% | 0 | 0 | — |
case-04 | pass→pass | 22,832 | 20,576 | -10% | 1 | 1 | 0% | 3,771 | 3,657 | -3% | 0 | 0 | — |
case-05 | pass→pass | 23,359 | 25,735 | +10% | 1 | 1 | 0% | 4,871 | 5,628 | +16% | 0 | 0 | — |
case-06 | pass→pass | 15,731 | 19,493 | +24% | 1 | 1 | 0% | 2,741 | 3,538 | +29% | 0 | 0 | — |
case-07 | pass→pass | 18,839 | 17,669 | -6% | 1 | 1 | 0% | 3,353 | 3,309 | -1% | 0 | 0 | — |
case-08 | fail→pass | 17,485 | 12,722 | -27% | 1 | 1 | 0% | 3,210 | 2,288 | -29% | 0 | 0 | — |
case-09 | pass→pass | 13,748 | 15,350 | +12% | 1 | 1 | 0% | 2,366 | 3,052 | +29% | 0 | 0 | — |
case-10 | pass→pass | 14,149 | 17,609 | +24% | 1 | 1 | 0% | 2,311 | 3,706 | +60% | 0 | 0 | — |
case-11 | pass→pass | 14,234 | 15,899 | +12% | 1 | 1 | 0% | 2,305 | 2,803 | +22% | 0 | 0 | — |
case-12 | fail→fail | 15,449 | 13,827 | -10% | 1 | 1 | 0% | 2,602 | 2,549 | -2% | 0 | 0 | — |
case-13 | fail→pass | 14,448 | 14,014 | -3% | 1 | 1 | 0% | 2,397 | 2,546 | +6% | 0 | 0 | — |
case-20 | pass→pass | 8,290 | 2,854 | -66% | 1 | 1 | 0% | 1,478 | 601 | -59% | 0 | 0 | — |
case-14 | pass→pass | 9,689 | 3,065 | -68% | 1 | 1 | 0% | 1,746 | 664 | -62% | 0 | 0 | — |
case-15 | pass→pass | 15,954 | 15,168 | -5% | 1 | 1 | 0% | 2,618 | 3,028 | +16% | 0 | 0 | — |
case-16 | pass→pass | 16,196 | 18,059 | +12% | 1 | 1 | 0% | 2,987 | 3,619 | +21% | 0 | 0 | — |
case-17 | fail→pass | 16,230 | 13,556 | -16% | 1 | 1 | 0% | 2,779 | 2,506 | -10% | 0 | 0 | — |
case-18 | pass→pass | 13,875 | 13,356 | -4% | 1 | 1 | 0% | 2,339 | 2,326 | -1% | 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 +23 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.