Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic for systematically extracting axes of variation from literature — identify how practitioners compare approaches.
.claude/skills/yogsoth-ai-axis-extraction/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -33% | 0% |
Systematically extract axes of variation from literature. Look for how authors compare methods, what trade-offs they discuss, what design choices they highlight.
<HARD-GATE> ≥2 candidate axes identified per invocation with independence assessment. </HARD-GATE>
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | axis-validation | SOP for validating that candidate axes are independent and meaningful. | | dimension-page-creation | SOP for creating a dimension page — documents an axis of variation with its values and semantics. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 16,510 | 17,845 | +8% | 1 | 1 | 0% | 2,652 | 3,170 | +20% | 0 | 0 | — |
case-02 | fail→pass | 19,105 | 18,916 | -1% | 1 | 1 | 0% | 3,004 | 3,327 | +11% | 0 | 0 | — |
case-03 | fail→pass | 19,998 | 15,131 | -24% | 1 | 1 | 0% | 3,147 | 2,707 | -14% | 0 | 0 | — |
case-04 | fail→pass | 25,836 | 10,765 | -58% | 1 | 1 | 0% | 1,967 | 1,919 | -2% | 0 | 0 | — |
case-05 | fail→pass | 18,508 | 9,750 | -47% | 1 | 1 | 0% | 2,649 | 1,769 | -33% | 0 | 0 | — |
case-06 | pass→pass | 15,461 | 10,257 | -34% | 1 | 1 | 0% | 2,254 | 1,850 | -18% | 0 | 0 | — |
case-07 | fail→pass | 8,393 | 9,528 | +14% | 1 | 1 | 0% | 1,328 | 1,788 | +35% | 0 | 0 | — |
case-08 | fail→pass | 10,114 | 1,715 | -83% | 1 | 1 | 0% | 1,476 | 550 | -63% | 0 | 0 | — |
case-09 | pass→pass | 26,748 | 1,811 | -93% | 1 | 1 | 0% | 1,971 | 527 | -73% | 0 | 0 | — |
case-10 | fail→pass | 16,611 | 1,972 | -88% | 1 | 1 | 0% | 2,564 | 614 | -76% | 0 | 0 | — |
case-11 | fail→pass | 3,357 | 9,515 | +183% | 1 | 1 | 0% | 602 | 1,803 | +200% | 0 | 0 | — |
case-12 | pass→pass | 7,284 | 9,756 | +34% | 1 | 1 | 0% | 1,090 | 1,601 | +47% | 0 | 0 | — |
case-13 | pass→pass | 5,480 | 7,910 | +44% | 1 | 1 | 0% | 951 | 1,665 | +75% | 0 | 0 | — |
case-19 | fail→pass | 5,640 | 8,891 | +58% | 1 | 1 | 0% | 918 | 1,612 | +76% | 0 | 0 | — |
case-14 | fail→fail | 2,455 | 2,171 | -12% | 1 | 1 | 0% | 394 | 646 | +64% | 0 | 0 | — |
case-15 | fail→pass | 11,631 | 7,917 | -32% | 1 | 1 | 0% | 1,829 | 1,562 | -15% | 0 | 0 | — |
case-16 | fail→pass | 13,805 | 13,310 | -4% | 1 | 1 | 0% | 2,206 | 2,483 | +13% | 0 | 0 | — |
case-17 | pass→pass | 15,532 | 11,783 | -24% | 1 | 1 | 0% | 2,307 | 2,054 | -11% | 0 | 0 | — |
case-18 | fail→pass | 10,632 | 7,046 | -34% | 1 | 1 | 0% | 1,731 | 1,606 | -7% | 0 | 0 | — |
case-20 | fail→pass | 17,668 | 18,360 | +4% | 1 | 1 | 0% | 2,859 | 3,332 | +17% | 0 | 0 | — |
case-21 | pass→pass | 12,849 | 3,651 | -72% | 1 | 1 | 0% | 1,967 | 860 | -56% | 0 | 0 | — |
case-22 | fail→pass | 13,621 | 12,744 | -6% | 1 | 1 | 0% | 2,090 | 2,165 | +4% | 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 +68 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.