Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Systematically catalog failure modes — generate edge cases, observe failures, cluster by mechanism, identify triggers and frequency.
.claude/skills/yogsoth-ai-failure-mode-analysis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 80% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 103% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -73% | 0% |
Build a comprehensive failure taxonomy for a method.
| Base SOP | Target | ±10% Range | |----------|--------|------------| | web-search | 40 | 36–44 | | web-research | 15 | 13–17 | | paper-overview | 40 | 36–44 | | paper-search | 25 | 22–28 | | paper-research | 10 | 9–11 |
<HARD-GATE>
| SOP | Done | Target | % |
|-----|------|--------|---|
| web-search | ? | 40 | ? |
| web-research | ? | 15 | ? |
| paper-overview | ? | 40 | ? |
| paper-search | ? | 25 | ? |
| paper-research | ? | 10 | ? |
Budget Gate: OPEN/CLOSED (>=80% required to exit)
</HARD-GATE>Import: web-search, web-research, paper-overview, paper-search, paper-research Subagent: edge-case-generation, failure-clustering, scaling-regime-detection
Generate systematic edge cases, observe failure patterns, cluster into failure modes, identify triggering conditions and frequency estimates.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | deep-insight-failure-mode-cataloging | Systematic failure mode cataloging — generate boundary inputs, observe failures, cluster by mechanism, identify triggers, estimate frequency. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | edge-case-generation | Systematically generate boundary inputs — boundary values, adversarial constructions, distribution shifts, rare combinations, scale extremes. | | failure-clustering | Group observed failures by mechanism (not symptom), identify common triggers per cluster, estimate frequency and severity. | | scaling-regime-detection | Detect regime changes in scaling behavior — breakpoints where behavior qualitatively shifts, mechanisms behind transitions. |
<!-- 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 | 37,998 | 50,774 | +34% | 1 | 1 | 0% | 5,295 | 6,738 | +27% | 0 | 0 | — |
case-02 | fail→pass | 31,977 | 56,305 | +76% | 1 | 1 | 0% | 4,247 | 6,744 | +59% | 0 | 0 | — |
case-03 | fail→fail | 35,521 | 12,994 | -63% | 1 | 1 | 0% | 4,899 | 1,286 | -74% | 0 | 0 | — |
case-04 | pass→pass | 20,555 | 23,795 | +16% | 1 | 1 | 0% | 3,862 | 5,054 | +31% | 0 | 0 | — |
case-05 | pass→pass | 13,431 | 19,325 | +44% | 1 | 1 | 0% | 2,663 | 4,216 | +58% | 0 | 0 | — |
case-06 | pass→pass | 17,650 | 34,066 | +93% | 1 | 1 | 0% | 2,994 | 6,706 | +124% | 0 | 0 | — |
case-07 | fail→fail | 23,691 | 7,420 | -69% | 1 | 1 | 0% | 3,247 | 1,605 | -51% | 0 | 0 | — |
case-08 | fail→pass | 47,285 | 48,445 | +2% | 1 | 1 | 0% | 3,729 | 6,714 | +80% | 0 | 0 | — |
case-09 | fail→pass | 42,730 | 35,259 | -17% | 1 | 1 | 0% | 2,888 | 5,858 | +103% | 0 | 0 | — |
case-10 | fail→pass | 40,515 | 8,355 | -79% | 1 | 1 | 0% | 5,918 | 1,571 | -73% | 0 | 0 | — |
case-11 | fail→pass | 26,681 | 41,526 | +56% | 1 | 1 | 0% | 3,642 | 6,710 | +84% | 0 | 0 | — |
case-12 | fail→pass | 48,838 | 78,326 | +60% | 1 | 1 | 0% | 1,386 | 6,706 | +384% | 0 | 0 | — |
case-13 | pass→fail | 22,754 | 7,505 | -67% | 1 | 1 | 0% | 3,457 | 1,456 | -58% | 0 | 0 | — |
case-14 | pass→fail | 27,489 | 5,689 | -79% | 1 | 1 | 0% | 3,710 | 1,284 | -65% | 0 | 0 | — |
case-15 | pass→pass | 26,229 | 42,918 | +64% | 1 | 1 | 0% | 3,671 | 6,707 | +83% | 0 | 0 | — |
case-16 | fail→pass | 27,452 | 41,833 | +52% | 1 | 1 | 0% | 4,112 | 6,705 | +63% | 0 | 0 | — |
case-17 | fail→pass | 29,072 | 39,217 | +35% | 1 | 1 | 0% | 4,216 | 6,706 | +59% | 0 | 0 | — |
case-18 | fail→pass | 22,936 | 43,598 | +90% | 1 | 1 | 0% | 3,615 | 6,708 | +86% | 0 | 0 | — |
case-19 | fail→fail | 25,976 | 9,152 | -65% | 1 | 1 | 0% | 3,484 | 1,083 | -69% | 0 | 0 | — |
case-20 | fail→pass | 29,481 | 40,868 | +39% | 1 | 1 | 0% | 4,238 | 6,712 | +58% | 0 | 0 | — |
case-21 | fail→fail | 16,375 | 37,510 | +129% | 1 | 1 | 0% | 2,613 | 6,160 | +136% | 0 | 0 | — |
case-22 | pass→fail | 29,599 | 7,586 | -74% | 1 | 1 | 0% | 3,885 | 896 | -77% | 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 18 counted toward the lift figure. The other 4 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 18 comparable cases. 3 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.