Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Systematic extraction, challenge, and sensitivity analysis of assumptions underlying a decision to identify load-bearing beliefs.
.claude/skills/yogsoth-ai-assumption-excavation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 13% | 0% |
A three-phase tactic that surfaces hidden assumptions, challenges each one adversarially, and maps which assumptions are load-bearing for the conclusion. Decisions often rest on unstated beliefs — this tactic makes them explicit and tests their strength.
| SOP | Phase | Purpose | |-----|-------|---------| | assumption-extraction | Extract | Surface hidden assumptions with confidence | | assumption-challenge | Challenge | Attack each assumption adversarially | | conclusion-sensitivity | Sensitivity | Map load-bearing assumptions |
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | conclusion-sensitivity | Map which assumptions are load-bearing by assessing how the conclusion changes if each assumption fails. | | convergence-assumption-challenge | Construct the strongest counter-argument against a specific assumption and propose alternatives. | | convergence-assumption-extraction | Systematically surface hidden assumptions underlying a decision with confidence levels. |
<!-- 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→fail | 21,315 | 25,882 | +21% | 1 | 1 | 0% | 3,086 | 4,360 | +41% | 0 | 0 | — |
case-02 | fail→pass | 27,640 | 22,883 | -17% | 1 | 1 | 0% | 4,079 | 3,830 | -6% | 0 | 0 | — |
case-03 | fail→pass | 21,241 | 23,973 | +13% | 1 | 1 | 0% | 3,216 | 3,979 | +24% | 0 | 0 | — |
case-04 | pass→pass | 18,964 | 21,897 | +15% | 1 | 1 | 0% | 2,676 | 3,707 | +39% | 0 | 0 | — |
case-05 | fail→pass | 20,975 | 24,489 | +17% | 1 | 1 | 0% | 3,024 | 4,055 | +34% | 0 | 0 | — |
case-06 | fail→pass | 21,135 | 22,046 | +4% | 1 | 1 | 0% | 2,774 | 3,793 | +37% | 0 | 0 | — |
case-07 | pass→pass | 21,244 | 20,779 | -2% | 1 | 1 | 0% | 3,132 | 3,648 | +16% | 0 | 0 | — |
case-08 | fail→pass | 21,296 | 22,060 | +4% | 1 | 1 | 0% | 3,214 | 3,636 | +13% | 0 | 0 | — |
case-09 | fail→pass | 23,300 | 20,150 | -14% | 1 | 1 | 0% | 3,337 | 3,461 | +4% | 0 | 0 | — |
case-10 | fail→pass | 23,368 | 20,206 | -14% | 1 | 1 | 0% | 3,221 | 3,556 | +10% | 0 | 0 | — |
case-11 | fail→pass | 18,011 | 21,395 | +19% | 1 | 1 | 0% | 2,607 | 3,692 | +42% | 0 | 0 | — |
case-12 | fail→pass | 23,569 | 23,760 | +1% | 1 | 1 | 0% | 3,377 | 3,920 | +16% | 0 | 0 | — |
case-13 | fail→pass | 19,149 | 21,480 | +12% | 1 | 1 | 0% | 2,764 | 3,549 | +28% | 0 | 0 | — |
case-14 | fail→pass | 22,086 | 19,775 | -10% | 1 | 1 | 0% | 3,009 | 3,461 | +15% | 0 | 0 | — |
case-15 | fail→pass | 21,426 | 20,303 | -5% | 1 | 1 | 0% | 2,807 | 3,377 | +20% | 0 | 0 | — |
case-16 | pass→pass | 16,379 | 22,160 | +35% | 1 | 1 | 0% | 2,453 | 3,628 | +48% | 0 | 0 | — |
case-17 | fail→pass | 21,867 | 22,800 | +4% | 1 | 1 | 0% | 3,135 | 3,900 | +24% | 0 | 0 | — |
case-18 | fail→pass | 18,365 | 20,978 | +14% | 1 | 1 | 0% | 2,655 | 3,485 | +31% | 0 | 0 | — |
case-19 | pass→fail | 14,621 | 24,546 | +68% | 1 | 1 | 0% | 2,108 | 4,122 | +96% | 0 | 0 | — |
case-20 | pass→fail | 18,264 | 35,925 | +97% | 1 | 1 | 0% | 3,000 | 6,136 | +105% | 0 | 0 | — |
case-21 | pass→pass | 14,713 | 17,075 | +16% | 1 | 1 | 0% | 2,264 | 3,032 | +34% | 0 | 0 | — |
case-22 | fail→pass | 18,950 | 16,896 | -11% | 1 | 1 | 0% | 2,777 | 3,097 | +12% | 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 +59 percentage points is the difference between those two pass rates over the 22 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.