Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build or extend a playable isometric action RPG in Three.js, React, or similar web technology. Use for game-loop architecture, camera and movement, zones, combat integration, content data, progression, saves, or production-ready vertical slices.
.claude/skills/mengto-build-isometric-arpg/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -15% | 0% |
Build one coherent playable loop before expanding scope: title/menu, character choice, movement, one encounter, reward, progression, and continuation.
Deliver each slice with a playable route and deterministic test state. Prefer this order:
Avoid adding a second system until the preceding slice has gameplay proof, automated coverage, and a stable save/load path.
Run focused unit tests, type/lint checks, production build, and a real playthrough in the repository-approved browser. Verify a fresh start, saved continuation, loss/retry, one complete objective, keyboard/mouse controls, and the intended mobile baseline. Report known baseline failures separately from new failures.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 20,777 | 18,571 | -11% | 1 | 1 | 0% | 3,683 | 3,822 | +4% | 0 | 0 | — |
case-02 | fail→fail | 25,707 | 32,464 | +26% | 1 | 1 | 0% | 4,427 | 6,350 | +43% | 0 | 0 | — |
case-03 | fail→fail | 24,124 | 12,908 | -46% | 1 | 1 | 0% | 4,086 | 2,618 | -36% | 0 | 0 | — |
case-04 | pass→pass | 11,696 | 5,066 | -57% | 1 | 1 | 0% | 1,835 | 1,225 | -33% | 0 | 0 | — |
case-05 | fail→pass | 13,354 | 7,208 | -46% | 1 | 1 | 0% | 2,055 | 1,660 | -19% | 0 | 0 | — |
case-06 | pass→pass | 12,736 | 6,325 | -50% | 1 | 1 | 0% | 2,053 | 1,522 | -26% | 0 | 0 | — |
case-07 | pass→pass | 17,306 | 10,650 | -38% | 1 | 1 | 0% | 3,151 | 2,430 | -23% | 0 | 0 | — |
case-08 | pass→pass | 11,535 | 6,958 | -40% | 1 | 1 | 0% | 1,975 | 1,585 | -20% | 0 | 0 | — |
case-09 | fail→pass | 11,710 | 8,987 | -23% | 1 | 1 | 0% | 1,966 | 1,844 | -6% | 0 | 0 | — |
case-10 | pass→pass | 10,672 | 8,297 | -22% | 1 | 1 | 0% | 1,960 | 1,835 | -6% | 0 | 0 | — |
case-11 | pass→pass | 9,303 | 4,808 | -48% | 1 | 1 | 0% | 1,581 | 1,274 | -19% | 0 | 0 | — |
case-12 | fail→pass | 7,299 | 6,427 | -12% | 1 | 1 | 0% | 1,215 | 1,522 | +25% | 0 | 0 | — |
case-13 | pass→pass | 16,626 | 14,845 | -11% | 1 | 1 | 0% | 2,815 | 3,004 | +7% | 0 | 0 | — |
case-14 | fail→pass | 16,598 | 14,750 | -11% | 1 | 1 | 0% | 2,603 | 2,873 | +10% | 0 | 0 | — |
case-15 | pass→pass | 16,575 | 8,657 | -48% | 1 | 1 | 0% | 2,715 | 1,905 | -30% | 0 | 0 | — |
case-16 | fail→pass | 12,418 | 8,099 | -35% | 1 | 1 | 0% | 2,088 | 1,773 | -15% | 0 | 0 | — |
case-17 | fail→pass | 12,763 | 8,936 | -30% | 1 | 1 | 0% | 2,237 | 1,960 | -12% | 0 | 0 | — |
case-18 | fail→pass | 15,863 | 4,212 | -73% | 1 | 1 | 0% | 2,561 | 1,120 | -56% | 0 | 0 | — |
case-19 | pass→pass | 17,846 | 5,043 | -72% | 1 | 1 | 0% | 2,939 | 1,349 | -54% | 0 | 0 | — |
case-20 | pass→pass | 3,065 | 3,215 | +5% | 1 | 1 | 0% | 506 | 1,001 | +98% | 0 | 0 | — |
case-21 | pass→pass | 2,637 | 2,793 | +6% | 1 | 1 | 0% | 407 | 860 | +111% | 0 | 0 | — |
case-22 | pass→pass | 4,002 | 3,533 | -12% | 1 | 1 | 0% | 748 | 1,094 | +46% | 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 +32 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.