Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic: Sequential perspective evaluation with divergence aggregation. Each agent evaluates from a distinct viewpoint, then disagreements are surfaced and resolved.
.claude/skills/yogsoth-ai-stress-test-perspective-rotation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -4% | 0% |
Sequential multi-perspective evaluation followed by divergence analysis and deliberation.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | confidence-calibration | Calibrates confidence scores based on debate progression. Determines whether to escalate, continue, or terminate based on cumulative evidence. | | debate-architect | Designs debate structure based on artifact type — selects attack vectors, assigns perspectives, determines escalation ladder, and configures round parameters. | | divergence-detection | Identifies agreement and disagreement patterns across multiple perspective evaluations. Maps consensus clusters and persistent divergence points. | | perspective-critic | Evaluates artifact from a specific assigned perspective. Produces assessment grounded in that viewpoint's values, priorities, and expertise. |
<!-- 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 | 113,384 | 39,831 | -65% | 1 | 1 | 0% | 6,758 | 5,373 | -20% | 0 | 0 | — |
case-02 | fail→fail | 49,705 | 38,378 | -23% | 1 | 1 | 0% | 7,140 | 5,117 | -28% | 0 | 0 | — |
case-03 | fail→pass | 33,743 | 32,120 | -5% | 1 | 1 | 0% | 4,376 | 4,670 | +7% | 0 | 0 | — |
case-04 | fail→fail | 48,019 | 26,141 | -46% | 1 | 1 | 0% | 8,256 | 4,799 | -42% | 0 | 0 | — |
case-05 | fail→pass | 45,590 | 8,340 | -82% | 1 | 1 | 0% | 2,598 | 913 | -65% | 0 | 0 | — |
case-15 | fail→pass | 18,504 | 16,857 | -9% | 1 | 1 | 0% | 1,948 | 2,275 | +17% | 0 | 0 | — |
case-06 | fail→pass | 15,275 | 8,454 | -45% | 1 | 1 | 0% | 1,521 | 1,013 | -33% | 0 | 0 | — |
case-07 | pass→pass | 13,415 | 15,768 | +18% | 1 | 1 | 0% | 1,910 | 2,154 | +13% | 0 | 0 | — |
case-08 | fail→pass | 32,081 | 12,231 | -62% | 1 | 1 | 0% | 1,828 | 1,762 | -4% | 0 | 0 | — |
case-09 | pass→pass | 18,097 | 12,049 | -33% | 1 | 1 | 0% | 2,671 | 1,796 | -33% | 0 | 0 | — |
case-10 | pass→pass | 15,323 | 8,367 | -45% | 1 | 1 | 0% | 1,547 | 920 | -41% | 0 | 0 | — |
case-11 | pass→pass | 8,737 | 8,049 | -8% | 1 | 1 | 0% | 1,299 | 982 | -24% | 0 | 0 | — |
case-12 | fail→pass | 13,096 | 8,253 | -37% | 1 | 1 | 0% | 1,139 | 1,027 | -10% | 0 | 0 | — |
case-13 | fail→pass | 28,569 | 2,442 | -91% | 1 | 1 | 0% | 3,881 | 963 | -75% | 0 | 0 | — |
case-14 | pass→pass | 19,068 | 10,909 | -43% | 1 | 1 | 0% | 2,061 | 2,190 | +6% | 0 | 0 | — |
case-16 | pass→pass | 13,094 | 19,869 | +52% | 1 | 1 | 0% | 1,959 | 2,703 | +38% | 0 | 0 | — |
case-17 | pass→pass | 23,377 | 23,621 | +1% | 1 | 1 | 0% | 2,643 | 3,129 | +18% | 0 | 0 | — |
case-18 | fail→pass | 15,708 | 8,374 | -47% | 1 | 1 | 0% | 1,664 | 1,050 | -37% | 0 | 0 | — |
case-19 | fail→pass | 19,586 | 10,057 | -49% | 1 | 1 | 0% | 2,132 | 1,120 | -47% | 0 | 0 | — |
case-20 | pass→pass | 24,410 | 39,624 | +62% | 1 | 1 | 0% | 2,297 | 4,534 | +97% | 0 | 0 | — |
case-21 | pass→fail | 18,280 | 22,290 | +22% | 1 | 1 | 0% | 2,561 | 3,979 | +55% | 0 | 0 | — |
case-22 | pass→pass | 23,009 | 33,733 | +47% | 1 | 1 | 0% | 2,925 | 4,900 | +68% | 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, and 21 counted toward the lift figure. The other 1 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 +36 percentage points is the difference between those two pass rates over the 21 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.