Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Design or implement responsive audio feedback for a Three.js or web game. Use for action sounds, combat layers, music states, spatial audio, mix priorities, mute controls, accessibility, mobile audio unlock, and audio performance.
.claude/skills/mengto-build-game-audio-feedback/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-20 | ✓→✗ | ▼ Worse | 9% | 0% |
Use audio to confirm player intent and combat state, not to add constant noise.
Assign distinct cues for input accepted, windup, contact, block, miss, damage, interrupt, pickup, warning, death, and objective completion. Define priority so critical feedback remains audible during crowded encounters.
Unlock audio from a user gesture, respect mute/volume settings, release completed nodes, cap simultaneous voices, and provide subtitles or visual equivalents for meaningful cues. Change music by state with controlled transitions rather than restarts.
Test first interaction, rapid actions, pause, tab/background return, mobile, muted state, and reduced motion/accessibility. Check memory and console health after a representative session.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 23,609 | 16,093 | -32% | 1 | 1 | 0% | 3,729 | 3,334 | -11% | 0 | 0 | — |
case-02 | fail→fail | 19,664 | 29,860 | +52% | 1 | 1 | 0% | 3,501 | 6,339 | +81% | 0 | 0 | — |
case-03 | fail→fail | 19,504 | 23,476 | +20% | 1 | 1 | 0% | 3,394 | 4,416 | +30% | 0 | 0 | — |
case-04 | pass→pass | 15,012 | 10,714 | -29% | 1 | 1 | 0% | 2,444 | 2,116 | -13% | 0 | 0 | — |
case-05 | pass→pass | 10,010 | 9,924 | -1% | 1 | 1 | 0% | 1,918 | 2,078 | +8% | 0 | 0 | — |
case-06 | pass→pass | 11,866 | 9,286 | -22% | 1 | 1 | 0% | 2,037 | 1,743 | -14% | 0 | 0 | — |
case-07 | pass→pass | 11,381 | 8,496 | -25% | 1 | 1 | 0% | 2,163 | 1,744 | -19% | 0 | 0 | — |
case-08 | fail→pass | 14,702 | 10,942 | -26% | 1 | 1 | 0% | 2,555 | 1,922 | -25% | 0 | 0 | — |
case-09 | pass→pass | 8,381 | 6,066 | -28% | 1 | 1 | 0% | 1,515 | 1,094 | -28% | 0 | 0 | — |
case-10 | pass→pass | 15,739 | 15,335 | -3% | 1 | 1 | 0% | 2,787 | 2,573 | -8% | 0 | 0 | — |
case-11 | pass→pass | 14,023 | 13,192 | -6% | 1 | 1 | 0% | 2,392 | 2,154 | -10% | 0 | 0 | — |
case-12 | fail→pass | 13,032 | 9,562 | -27% | 1 | 1 | 0% | 2,158 | 1,597 | -26% | 0 | 0 | — |
case-13 | fail→pass | 11,073 | 4,949 | -55% | 1 | 1 | 0% | 1,771 | 948 | -46% | 0 | 0 | — |
case-14 | pass→pass | 17,419 | 11,707 | -33% | 1 | 1 | 0% | 2,744 | 2,164 | -21% | 0 | 0 | — |
case-15 | pass→pass | 13,255 | 9,363 | -29% | 1 | 1 | 0% | 2,211 | 1,653 | -25% | 0 | 0 | — |
case-16 | fail→fail | 14,182 | 14,518 | +2% | 1 | 1 | 0% | 2,458 | 2,568 | +4% | 0 | 0 | — |
case-17 | pass→pass | 14,232 | 12,309 | -14% | 1 | 1 | 0% | 2,267 | 2,089 | -8% | 0 | 0 | — |
case-18 | pass→pass | 14,489 | 13,370 | -8% | 1 | 1 | 0% | 2,529 | 2,371 | -6% | 0 | 0 | — |
case-19 | pass→pass | 13,792 | 11,891 | -14% | 1 | 1 | 0% | 2,366 | 2,243 | -5% | 0 | 0 | — |
case-20 | pass→fail | 17,704 | 20,374 | +15% | 1 | 1 | 0% | 3,810 | 4,149 | +9% | 0 | 0 | — |
case-21 | pass→pass | 14,459 | 16,372 | +13% | 1 | 1 | 0% | 3,261 | 3,683 | +13% | 0 | 0 | — |
case-22 | pass→pass | 15,037 | 11,135 | -26% | 1 | 1 | 0% | 2,781 | 2,145 | -23% | 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.