Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Split an oversized skill, CLAUDE.md, or spec into an entry file plus files that load only when they're needed.
.claude/skills/neeeophytee-progressive-disclosure/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-03 | ✓→✓ | = Same ✓ | 28% | 0% |
| case-04 | ✓→✓ | = Same ✓ | -7% | 0% |
A long instruction file is paid for on every single turn, including the turns that need none of it. The fix is not deletion — the material is real — but placement: keep what every run needs in the entry file, and move what only some runs need behind a pointer that fires when it's relevant. This skill restructures one artifact. To find out which artifacts need it, audit first.
verification.md, glossary.md — so the name alone tells the reader when to open it.migrations.md before planning." A bare link at the bottom of a file is not a pointer.SKILL.md before relying on them; installers differ, and a pointer to a file that didn't travel is a broken skill.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 19,962 | 24,792 | +24% | 1 | 1 | 0% | 3,325 | 3,938 | +18% | 0 | 0 | — |
case-02 | fail→pass | 10,240 | 7,377 | -28% | 1 | 1 | 0% | 1,717 | 1,752 | +2% | 0 | 0 | — |
case-03 | pass→pass | 5,869 | 4,496 | -23% | 1 | 1 | 0% | 968 | 1,241 | +28% | 0 | 0 | — |
case-04 | pass→pass | 9,880 | 5,364 | -46% | 1 | 1 | 0% | 1,466 | 1,366 | -7% | 0 | 0 | — |
case-05 | pass→pass | 6,258 | 3,403 | -46% | 1 | 1 | 0% | 907 | 1,021 | +13% | 0 | 0 | — |
case-06 | pass→pass | 14,018 | 4,726 | -66% | 1 | 1 | 0% | 2,158 | 1,278 | -41% | 0 | 0 | — |
case-07 | fail→pass | 6,755 | 2,327 | -66% | 1 | 1 | 0% | 956 | 895 | -6% | 0 | 0 | — |
case-08 | fail→fail | 6,945 | 4,061 | -42% | 1 | 1 | 0% | 1,021 | 1,185 | +16% | 0 | 0 | — |
case-09 | pass→pass | 9,663 | 3,523 | -64% | 1 | 1 | 0% | 1,500 | 1,060 | -29% | 0 | 0 | — |
case-10 | pass→pass | 6,286 | 3,662 | -42% | 1 | 1 | 0% | 1,014 | 1,115 | +10% | 0 | 0 | — |
case-11 | pass→pass | 10,054 | 7,207 | -28% | 1 | 1 | 0% | 1,477 | 1,669 | +13% | 0 | 0 | — |
case-12 | pass→pass | 9,968 | 3,747 | -62% | 1 | 1 | 0% | 1,544 | 1,193 | -23% | 0 | 0 | — |
case-13 | pass→pass | 12,272 | 2,536 | -79% | 1 | 1 | 0% | 1,708 | 933 | -45% | 0 | 0 | — |
case-14 | pass→pass | 6,571 | 4,084 | -38% | 1 | 1 | 0% | 996 | 1,187 | +19% | 0 | 0 | — |
case-15 | pass→pass | 11,886 | 9,485 | -20% | 1 | 1 | 0% | 1,295 | 2,027 | +57% | 0 | 0 | — |
case-16 | fail→fail | 9,101 | 2,561 | -72% | 1 | 1 | 0% | 1,319 | 872 | -34% | 0 | 0 | — |
case-17 | pass→pass | 10,774 | 5,096 | -53% | 1 | 1 | 0% | 1,608 | 1,368 | -15% | 0 | 0 | — |
case-18 | pass→pass | 10,402 | 6,124 | -41% | 1 | 1 | 0% | 1,638 | 1,449 | -12% | 0 | 0 | — |
case-19 | pass→pass | 11,002 | 8,291 | -25% | 1 | 1 | 0% | 1,591 | 1,752 | +10% | 0 | 0 | — |
case-20 | pass→pass | 12,439 | 9,979 | -20% | 1 | 1 | 0% | 1,974 | 2,068 | +5% | 0 | 0 | — |
case-21 | fail→fail | 8,415 | 3,632 | -57% | 1 | 1 | 0% | 1,351 | 1,156 | -14% | 0 | 0 | — |
case-22 | pass→pass | 8,912 | 2,258 | -75% | 1 | 1 | 0% | 1,320 | 942 | -29% | 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.
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.