Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build truthful Vesperfall asset-library review pairs from transparent PNG references and live Three.js, FBX, or img2threejs models. Use when adding a character, enemy, prop, or equipment asset to the Vesperfall catalog; creating a card-PNG plus inspector-model treatment; exposing an isolated model or moveset route; or standardizing provenance, grounding, tests, and Codex-browser validation for game art review.
.claude/skills/mengto-build-vesperfall-review-assets/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -15% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -49% | 0% |
Create one reviewable asset pair without confusing catalog media, imported files, procedural factories, and future contracts.
$img2threejs only when reconstructing from an image in code. Never relabel imported geometry as img2threejs.public/asset-catalog/<category>/.userData: source paths or factory, action count, sockets, and review-only status.scripts/validate_pair.py <png> <model-source>.git diff --check.Read references/catalog-contract.md when changing catalog types, provenance language, or review-route behavior.
Report:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 25,960 | 3,831 | -85% | 1 | 1 | 0% | 6,197 | 972 | -84% | 0 | 0 | — |
case-02 | fail→fail | 26,976 | 3,220 | -88% | 1 | 1 | 0% | 6,196 | 828 | -87% | 0 | 0 | — |
case-03 | fail→fail | 4,211 | 5,353 | +27% | 1 | 1 | 0% | 188 | 858 | +356% | 0 | 0 | — |
case-04 | pass→pass | 9,501 | 4,181 | -56% | 1 | 1 | 0% | 1,629 | 1,395 | -14% | 0 | 0 | — |
case-05 | fail→pass | 8,509 | 3,477 | -59% | 1 | 1 | 0% | 1,614 | 1,322 | -18% | 0 | 0 | — |
case-06 | pass→pass | 11,504 | 4,386 | -62% | 1 | 1 | 0% | 1,824 | 1,374 | -25% | 0 | 0 | — |
case-07 | fail→pass | 11,272 | 5,143 | -54% | 1 | 1 | 0% | 1,926 | 1,634 | -15% | 0 | 0 | — |
case-08 | pass→pass | 11,998 | 3,345 | -72% | 1 | 1 | 0% | 2,053 | 1,211 | -41% | 0 | 0 | — |
case-09 | fail→pass | 9,923 | 2,023 | -80% | 1 | 1 | 0% | 1,726 | 1,018 | -41% | 0 | 0 | — |
case-10 | fail→pass | 12,633 | 3,007 | -76% | 1 | 1 | 0% | 2,168 | 1,129 | -48% | 0 | 0 | — |
case-11 | pass→pass | 7,001 | 2,372 | -66% | 1 | 1 | 0% | 1,128 | 1,035 | -8% | 0 | 0 | — |
case-12 | pass→pass | 8,377 | 3,733 | -55% | 1 | 1 | 0% | 1,293 | 1,336 | +3% | 0 | 0 | — |
case-13 | pass→pass | 10,873 | 4,732 | -56% | 1 | 1 | 0% | 1,675 | 1,410 | -16% | 0 | 0 | — |
case-14 | fail→pass | 16,046 | 3,837 | -76% | 1 | 1 | 0% | 2,491 | 1,276 | -49% | 0 | 0 | — |
case-15 | pass→pass | 12,736 | 6,617 | -48% | 1 | 1 | 0% | 1,939 | 2,010 | +4% | 0 | 0 | — |
case-16 | pass→pass | 15,921 | 7,685 | -52% | 1 | 1 | 0% | 3,094 | 2,157 | -30% | 0 | 0 | — |
case-17 | fail→pass | 11,935 | 3,076 | -74% | 1 | 1 | 0% | 1,815 | 1,213 | -33% | 0 | 0 | — |
case-18 | pass→pass | 7,315 | 1,907 | -74% | 1 | 1 | 0% | 1,236 | 1,050 | -15% | 0 | 0 | — |
case-19 | fail→fail | 7,391 | 1,385 | -81% | 1 | 1 | 0% | 1,306 | 925 | -29% | 0 | 0 | — |
case-20 | pass→fail | 20,702 | 3,568 | -83% | 1 | 1 | 0% | 4,088 | 818 | -80% | 0 | 0 | — |
case-21 | pass→pass | 17,348 | 25,523 | +47% | 1 | 1 | 0% | 3,297 | 6,516 | +98% | 0 | 0 | — |
case-22 | pass→fail | 10,491 | 11,893 | +13% | 1 | 1 | 0% | 2,204 | 3,247 | +47% | 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, and 18 counted toward the lift figure. The other 4 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +18 percentage points is the difference between those two pass rates over the 18 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.