Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →5 humanities skills. Trigger: textual analysis, archival research, digital humanities, philosophy. Design: digital tools and qualitative methods for humanities scholarship.
.claude/skills/brycewang-stanford-humanities-skills/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 54% | 0% |
| case-23 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-04 | ✓→✗ | ▼ Worse | -87% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -77% | 0% |
Select the skill matching the user's need, then read its SKILL.md.
| Skill | Description | |-------|-------------| | digital-humanities-guide | Computational methods for humanities research including text mining and netwo... | | ethical-philosophy-guide | Applied ethics research methods and major ethical frameworks | | history-research-guide | Historical research from primary sources to scholarly analysis | | philosophy-research-guide | Research methods and analytical frameworks for philosophical inquiry and scho... | | political-history-guide | Chinese and European political struggle history and comparative analysis |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 14,352 | 13,299 | -7% | 1 | 1 | 0% | 2,239 | 466 | -79% | 0 | 0 | — |
case-02 | fail→fail | 6,671 | 14,318 | +115% | 1 | 1 | 0% | 374 | 553 | +48% | 0 | 0 | — |
case-03 | fail→fail | 20,576 | 11,898 | -42% | 1 | 1 | 0% | 3,112 | 611 | -80% | 0 | 0 | — |
case-04 | pass→fail | 18,569 | 3,546 | -81% | 1 | 1 | 0% | 3,421 | 456 | -87% | 0 | 0 | — |
case-05 | fail→pass | 32,579 | 19,584 | -40% | 1 | 1 | 0% | 3,126 | 2,812 | -10% | 0 | 0 | — |
case-06 | fail→fail | 20,337 | 3,551 | -83% | 1 | 1 | 0% | 2,437 | 520 | -79% | 0 | 0 | — |
case-07 | fail→fail | 20,843 | 9,951 | -52% | 1 | 1 | 0% | 2,497 | 537 | -78% | 0 | 0 | — |
case-08 | pass→pass | 31,937 | 23,418 | -27% | 1 | 1 | 0% | 4,248 | 1,721 | -59% | 0 | 0 | — |
case-09 | pass→fail | 15,020 | 16,224 | +8% | 1 | 1 | 0% | 2,342 | 527 | -77% | 0 | 0 | — |
case-10 | fail→fail | 20,937 | 15,109 | -28% | 1 | 1 | 0% | 3,342 | 537 | -84% | 0 | 0 | — |
case-11 | pass→fail | 14,789 | 5,183 | -65% | 1 | 1 | 0% | 2,314 | 506 | -78% | 0 | 0 | — |
case-12 | pass→fail | 27,153 | 8,508 | -69% | 1 | 1 | 0% | 2,947 | 459 | -84% | 0 | 0 | — |
case-13 | fail→fail | 24,771 | 4,815 | -81% | 1 | 1 | 0% | 2,926 | 430 | -85% | 0 | 0 | — |
case-14 | pass→fail | 26,379 | 9,610 | -64% | 1 | 1 | 0% | 3,625 | 533 | -85% | 0 | 0 | — |
case-15 | fail→fail | 10,740 | 15,545 | +45% | 1 | 1 | 0% | 391 | 566 | +45% | 0 | 0 | — |
case-16 | fail→fail | 24,347 | 5,457 | -78% | 1 | 1 | 0% | 2,890 | 565 | -80% | 0 | 0 | — |
case-17 | pass→fail | 20,225 | 4,144 | -80% | 1 | 1 | 0% | 2,416 | 432 | -82% | 0 | 0 | — |
case-18 | fail→fail | 25,051 | 15,135 | -40% | 1 | 1 | 0% | 2,611 | 511 | -80% | 0 | 0 | — |
case-19 | pass→fail | 29,545 | 16,126 | -45% | 1 | 1 | 0% | 3,163 | 559 | -82% | 0 | 0 | — |
case-20 | pass→fail | 24,081 | 16,863 | -30% | 1 | 1 | 0% | 2,940 | 538 | -82% | 0 | 0 | — |
case-21 | fail→pass | 23,246 | 30,889 | +33% | 1 | 1 | 0% | 2,425 | 3,729 | +54% | 0 | 0 | — |
case-22 | fail→fail | 21,375 | 4,925 | -77% | 1 | 1 | 0% | 3,364 | 627 | -81% | 0 | 0 | — |
case-23 | fail→pass | 26,700 | 33,577 | +26% | 1 | 1 | 0% | 3,720 | 5,085 | +37% | 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. 23 cases were attempted, and 4 counted toward the lift figure. The other 19 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 -22 percentage points is the difference between those two pass rates over the 4 comparable cases. 9 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.