Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Design, implement, tune, or test Three.js action-game encounters. Use for arena layout, enemy composition, spawn pacing, objectives, boss phases, reward cadence, encounter fixtures, and difficulty validation.
.claude/skills/mengto-design-game-encounters/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -2% | 0% |
Design encounters as decisions, not actor counts.
Define objective, available space, enemy roles, spawn timing, hazards, player resources, exits, failure recovery, and reward. Add one pressure source at a time; require each archetype to create a distinct response.
Protect readable paths, telegraphs, camera sight lines, and recovery windows. Cap simultaneous committed attackers and avoid offscreen damage, unavoidable chains, or encounter resets that duplicate rewards.
Create deterministic starts for low resources, each wave, boss phase, victory, and death/retry. Test desktop and mobile at the real camera distance, then adjust composition from observed decisions rather than raw completion time.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 21,115 | 17,929 | -15% | 1 | 1 | 0% | 3,368 | 2,895 | -14% | 0 | 0 | — |
case-02 | fail→fail | 19,443 | 17,501 | -10% | 1 | 1 | 0% | 3,129 | 3,170 | +1% | 0 | 0 | — |
case-03 | fail→pass | 20,032 | 66,712 | +233% | 1 | 1 | 0% | 3,234 | 3,465 | +7% | 0 | 0 | — |
case-04 | fail→pass | 19,089 | 14,566 | -24% | 1 | 1 | 0% | 3,035 | 2,572 | -15% | 0 | 0 | — |
case-05 | pass→pass | 14,926 | 34,926 | +134% | 1 | 1 | 0% | 2,432 | 2,511 | +3% | 0 | 0 | — |
case-06 | fail→fail | 19,341 | 14,318 | -26% | 1 | 1 | 0% | 2,865 | 2,355 | -18% | 0 | 0 | — |
case-07 | fail→pass | 20,440 | 15,204 | -26% | 1 | 1 | 0% | 3,143 | 2,606 | -17% | 0 | 0 | — |
case-08 | fail→pass | 20,308 | 15,672 | -23% | 1 | 1 | 0% | 2,826 | 2,775 | -2% | 0 | 0 | — |
case-09 | pass→pass | 16,869 | 23,157 | +37% | 1 | 1 | 0% | 2,583 | 3,817 | +48% | 0 | 0 | — |
case-10 | fail→pass | 18,442 | 11,228 | -39% | 1 | 1 | 0% | 2,869 | 1,887 | -34% | 0 | 0 | — |
case-11 | pass→fail | 17,255 | 12,939 | -25% | 1 | 1 | 0% | 2,675 | 2,115 | -21% | 0 | 0 | — |
case-12 | fail→pass | 18,177 | 16,605 | -9% | 1 | 1 | 0% | 3,191 | 2,813 | -12% | 0 | 0 | — |
case-13 | pass→pass | 19,723 | 15,538 | -21% | 1 | 1 | 0% | 2,752 | 2,504 | -9% | 0 | 0 | — |
case-14 | fail→pass | 13,786 | 20,665 | +50% | 1 | 1 | 0% | 2,553 | 3,527 | +38% | 0 | 0 | — |
case-15 | fail→pass | 17,250 | 17,395 | +1% | 1 | 1 | 0% | 2,788 | 2,881 | +3% | 0 | 0 | — |
case-16 | fail→pass | 21,081 | 21,579 | +2% | 1 | 1 | 0% | 3,198 | 3,473 | +9% | 0 | 0 | — |
case-17 | fail→fail | 18,887 | 11,363 | -40% | 1 | 1 | 0% | 3,010 | 1,898 | -37% | 0 | 0 | — |
case-18 | pass→pass | 13,986 | 15,322 | +10% | 1 | 1 | 0% | 2,145 | 2,551 | +19% | 0 | 0 | — |
case-19 | pass→fail | 18,732 | 21,410 | +14% | 1 | 1 | 0% | 3,142 | 3,017 | -4% | 0 | 0 | — |
case-20 | pass→pass | 18,266 | 21,040 | +15% | 1 | 1 | 0% | 3,033 | 3,415 | +13% | 0 | 0 | — |
case-21 | pass→pass | 14,359 | 10,545 | -27% | 1 | 1 | 0% | 2,597 | 2,066 | -20% | 0 | 0 | — |
case-22 | pass→pass | 24,458 | 24,689 | +1% | 1 | 1 | 0% | 4,746 | 4,988 | +5% | 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 +36 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are 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.