Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Assign distinct perspectives to attack a decision from multiple angles, then synthesize findings into a unified assessment.
.claude/skills/yogsoth-ai-multi-perspective-attack/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -48% | 0% |
A structured approach to attacking a decision from multiple distinct viewpoints simultaneously. Each perspective brings different values, concerns, and failure modes — ensuring blind spots from any single viewpoint are exposed.
| SOP | Phase | Purpose | |-----|-------|---------| | perspective-assignment | Assign | Define perspective briefs | | perspective-attack | Attack | Execute attack from each perspective | | steel-manning-synthesis | Synthesize | Unify findings into verdict |
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | perspective-assignment | Define distinct stakeholder or analytical perspectives with their values, concerns, and evaluation criteria. | | perspective-attack | Attack a decision from a specific assigned perspective, producing rated arguments and constructive alternatives. | | steel-manning-synthesis | Synthesize all attacks and verdicts into a final unified assessment with surviving concerns and recommended modifications. |
<!-- 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 | 47,072 | 19,132 | -59% | 1 | 1 | 0% | 2,779 | 2,624 | -6% | 0 | 0 | — |
case-02 | fail→pass | 18,122 | 21,365 | +18% | 1 | 1 | 0% | 2,491 | 2,943 | +18% | 0 | 0 | — |
case-03 | fail→fail | 20,097 | 26,989 | +34% | 1 | 1 | 0% | 2,015 | 2,886 | +43% | 0 | 0 | — |
case-04 | fail→pass | 22,800 | 17,949 | -21% | 1 | 1 | 0% | 2,088 | 2,385 | +14% | 0 | 0 | — |
case-05 | fail→fail | 27,211 | 17,027 | -37% | 1 | 1 | 0% | 1,801 | 3,273 | +82% | 0 | 0 | — |
case-06 | fail→fail | 12,532 | 32,061 | +156% | 1 | 1 | 0% | 1,042 | 2,739 | +163% | 0 | 0 | — |
case-07 | fail→pass | 20,520 | 32,739 | +60% | 1 | 1 | 0% | 2,214 | 1,607 | -27% | 0 | 0 | — |
case-08 | fail→pass | 19,016 | 9,268 | -51% | 1 | 1 | 0% | 2,272 | 1,174 | -48% | 0 | 0 | — |
case-09 | fail→pass | 20,293 | 8,880 | -56% | 1 | 1 | 0% | 2,276 | 1,031 | -55% | 0 | 0 | — |
case-10 | fail→fail | 28,068 | 17,416 | -38% | 1 | 1 | 0% | 1,522 | 2,144 | +41% | 0 | 0 | — |
case-11 | fail→fail | 18,638 | 37,118 | +99% | 1 | 1 | 0% | 1,831 | 2,947 | +61% | 0 | 0 | — |
case-12 | pass→pass | 50,926 | 26,220 | -49% | 1 | 1 | 0% | 2,041 | 3,601 | +76% | 0 | 0 | — |
case-13 | pass→pass | 40,339 | 43,935 | +9% | 1 | 1 | 0% | 2,789 | 3,368 | +21% | 0 | 0 | — |
case-14 | fail→pass | 19,619 | 54,305 | +177% | 1 | 1 | 0% | 2,034 | 4,226 | +108% | 0 | 0 | — |
case-15 | fail→fail | 25,457 | 25,832 | +1% | 1 | 1 | 0% | 2,997 | 3,510 | +17% | 0 | 0 | — |
case-16 | fail→pass | 25,954 | 29,527 | +14% | 1 | 1 | 0% | 3,009 | 4,163 | +38% | 0 | 0 | — |
case-17 | fail→fail | 11,489 | 15,866 | +38% | 1 | 1 | 0% | 1,096 | 1,759 | +60% | 0 | 0 | — |
case-18 | fail→fail | 16,693 | 24,625 | +48% | 1 | 1 | 0% | 1,736 | 3,550 | +104% | 0 | 0 | — |
case-19 | fail→fail | 8,563 | 3,173 | -63% | 1 | 1 | 0% | 428 | 874 | +104% | 0 | 0 | — |
case-20 | pass→pass | 18,165 | 21,644 | +19% | 1 | 1 | 0% | 2,099 | 3,900 | +86% | 0 | 0 | — |
case-21 | pass→fail | 23,707 | 24,469 | +3% | 1 | 1 | 0% | 2,955 | 3,616 | +22% | 0 | 0 | — |
case-22 | pass→pass | 24,091 | 39,870 | +65% | 1 | 1 | 0% | 3,293 | 6,637 | +102% | 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 +32 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.