Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Slightly drifted variant of a baking recipe used to validate that near-duplicate detection clusters skills that share most of their content but not all of it.
.claude/skills/galelmalah-near-recipe/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -77% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -46% | 0% |
The body here intentionally overlaps heavily with another skill of the same name so that MinHash clusters them as a near-duplicate while the content hash stays distinct.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | fail→fail | 2,393 | 1,589 | -34% | 1 | 1 | 0% | 368 | 412 | +12% | 0 | 0 | — |
case-01 | fail→fail | 13,367 | 4,920 | -63% | 1 | 1 | 0% | 1,953 | 1,086 | -44% | 0 | 0 | — |
case-02 | fail→pass | 12,754 | 4,057 | -68% | 1 | 1 | 0% | 2,214 | 820 | -63% | 0 | 0 | — |
case-03 | fail→pass | 12,752 | 3,690 | -71% | 1 | 1 | 0% | 2,376 | 804 | -66% | 0 | 0 | — |
case-04 | pass→pass | 5,335 | 6,106 | +14% | 1 | 1 | 0% | 916 | 1,027 | +12% | 0 | 0 | — |
case-05 | pass→pass | 7,697 | 4,912 | -36% | 1 | 1 | 0% | 1,516 | 1,021 | -33% | 0 | 0 | — |
case-06 | pass→pass | 4,812 | 3,822 | -21% | 1 | 1 | 0% | 919 | 895 | -3% | 0 | 0 | — |
case-07 | fail→pass | 10,593 | 1,931 | -82% | 1 | 1 | 0% | 2,076 | 471 | -77% | 0 | 0 | — |
case-08 | fail→fail | 7,969 | 3,282 | -59% | 1 | 1 | 0% | 1,311 | 642 | -51% | 0 | 0 | — |
case-09 | fail→pass | 11,134 | 2,995 | -73% | 1 | 1 | 0% | 1,794 | 594 | -67% | 0 | 0 | — |
case-10 | fail→pass | 8,092 | 3,533 | -56% | 1 | 1 | 0% | 1,248 | 679 | -46% | 0 | 0 | — |
case-11 | pass→pass | 12,917 | 5,866 | -55% | 1 | 1 | 0% | 1,991 | 993 | -50% | 0 | 0 | — |
case-12 | pass→pass | 9,490 | 3,273 | -66% | 1 | 1 | 0% | 1,450 | 723 | -50% | 0 | 0 | — |
case-14 | fail→pass | 3,014 | 1,570 | -48% | 1 | 1 | 0% | 418 | 342 | -18% | 0 | 0 | — |
case-15 | pass→pass | 7,426 | 2,983 | -60% | 1 | 1 | 0% | 1,216 | 578 | -52% | 0 | 0 | — |
case-16 | fail→pass | 1,605 | 1,647 | +3% | 1 | 1 | 0% | 193 | 416 | +116% | 0 | 0 | — |
case-17 | pass→pass | 8,782 | 1,941 | -78% | 1 | 1 | 0% | 1,381 | 458 | -67% | 0 | 0 | — |
case-18 | pass→pass | 7,441 | 2,745 | -63% | 1 | 1 | 0% | 1,355 | 585 | -57% | 0 | 0 | — |
case-19 | pass→pass | 7,025 | 1,891 | -73% | 1 | 1 | 0% | 1,207 | 457 | -62% | 0 | 0 | — |
case-20 | pass→pass | 16,057 | 1,566 | -90% | 1 | 1 | 0% | 1,897 | 321 | -83% | 0 | 0 | — |
case-21 | fail→pass | 1,403 | 830 | -41% | 1 | 1 | 0% | 164 | 212 | +29% | 0 | 0 | — |
case-22 | fail→pass | 11,066 | 1,802 | -84% | 1 | 1 | 0% | 1,984 | 429 | -78% | 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 +41 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | +50% |
Other measured skills in the registry, with their headline benchmark lift.