Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Produce a structured disagreement map showing clusters, arguments, and fault lines.
.claude/skills/yogsoth-ai-disagreement-visualization/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 448% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 14% | 0% |
Produce a structured disagreement map that shows the topology of disagreement: which clusters exist, what arguments support each, where the fault lines lie, and what type of disagreement separates them.
Spawn a subagent that takes clusters and their extracted arguments, then produces a structured map showing relationships, tensions, and potential bridges.
Output MUST contain: disagreement_map with clusters, fault_lines (at least 1 if multiple clusters exist), and fault_line_types. Map must cover ALL input clusters.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. Used by SOPs that declare execution: subagent. |
<!-- 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 | 18,092 | 25,238 | +39% | 1 | 1 | 0% | 3,093 | 3,595 | +16% | 0 | 0 | — |
case-02 | fail→pass | 13,724 | 16,346 | +19% | 1 | 1 | 0% | 2,287 | 3,120 | +36% | 0 | 0 | — |
case-03 | fail→pass | 21,649 | 27,377 | +26% | 1 | 1 | 0% | 3,464 | 4,606 | +33% | 0 | 0 | — |
case-04 | pass→fail | 10,587 | 23,209 | +119% | 1 | 1 | 0% | 1,588 | 4,336 | +173% | 0 | 0 | — |
case-05 | fail→fail | 3,536 | 4,371 | +24% | 1 | 1 | 0% | 601 | 968 | +61% | 0 | 0 | — |
case-06 | fail→pass | 2,942 | 12,045 | +309% | 1 | 1 | 0% | 424 | 2,323 | +448% | 0 | 0 | — |
case-07 | fail→pass | 19,066 | 19,016 | -0% | 1 | 1 | 0% | 2,995 | 3,412 | +14% | 0 | 0 | — |
case-08 | fail→pass | 15,514 | 20,777 | +34% | 1 | 1 | 0% | 2,567 | 3,911 | +52% | 0 | 0 | — |
case-09 | pass→pass | 17,973 | 13,942 | -22% | 1 | 1 | 0% | 2,780 | 2,652 | -5% | 0 | 0 | — |
case-10 | fail→fail | 27,800 | 29,392 | +6% | 1 | 1 | 0% | 4,576 | 5,354 | +17% | 0 | 0 | — |
case-11 | fail→fail | 19,069 | 7,549 | -60% | 1 | 1 | 0% | 3,140 | 1,011 | -68% | 0 | 0 | — |
case-12 | pass→pass | 18,565 | 15,280 | -18% | 1 | 1 | 0% | 3,118 | 2,880 | -8% | 0 | 0 | — |
case-13 | fail→pass | 16,492 | 15,493 | -6% | 1 | 1 | 0% | 2,660 | 3,010 | +13% | 0 | 0 | — |
case-19 | fail→pass | 18,941 | 29,779 | +57% | 1 | 1 | 0% | 2,794 | 3,820 | +37% | 0 | 0 | — |
case-14 | fail→pass | 14,736 | 20,280 | +38% | 1 | 1 | 0% | 2,266 | 3,630 | +60% | 0 | 0 | — |
case-15 | pass→pass | 16,879 | 18,638 | +10% | 1 | 1 | 0% | 2,565 | 3,285 | +28% | 0 | 0 | — |
case-16 | fail→pass | 21,767 | 25,838 | +19% | 1 | 1 | 0% | 3,483 | 3,699 | +6% | 0 | 0 | — |
case-17 | fail→pass | 21,756 | 22,863 | +5% | 1 | 1 | 0% | 3,340 | 2,725 | -18% | 0 | 0 | — |
case-18 | fail→pass | 19,672 | 26,624 | +35% | 1 | 1 | 0% | 3,083 | 4,816 | +56% | 0 | 0 | — |
case-20 | fail→pass | 21,964 | 20,834 | -5% | 1 | 1 | 0% | 3,328 | 3,715 | +12% | 0 | 0 | — |
case-21 | fail→pass | 17,492 | 18,497 | +6% | 1 | 1 | 0% | 2,861 | 3,357 | +17% | 0 | 0 | — |
case-22 | fail→pass | 19,390 | 24,264 | +25% | 1 | 1 | 0% | 2,915 | 4,163 | +43% | 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 +64 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.