Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Identify natural opinion clusters from collected judgments and characterize each cluster.
.claude/skills/yogsoth-ai-cluster-analysis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 214% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 102% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 264% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 52% | 0% |
Identify natural groupings of similar positions within the collected judgments. Characterize each cluster by its central position, shared reasoning patterns, and distinguishing features.
Spawn a subagent that analyzes the judgments for similarity patterns, groups them into coherent clusters, and provides characterization of each cluster.
Output MUST contain: at least 2 clusters (if genuine disagreement exists), each with cluster_id, position_summary, member_count, and characterization. If all judgments agree, output 1 cluster with a note.
<!-- 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 | 5,705 | 15,066 | +164% | 1 | 1 | 0% | 1,022 | 3,214 | +214% | 0 | 0 | — |
case-02 | fail→pass | 6,926 | 11,968 | +73% | 1 | 1 | 0% | 1,226 | 2,472 | +102% | 0 | 0 | — |
case-03 | fail→fail | 2,085 | 10,763 | +416% | 1 | 1 | 0% | 385 | 2,308 | +499% | 0 | 0 | — |
case-04 | fail→pass | 15,207 | 9,180 | -40% | 1 | 1 | 0% | 2,521 | 1,958 | -22% | 0 | 0 | — |
case-05 | fail→fail | 3,594 | 16,536 | +360% | 1 | 1 | 0% | 551 | 3,161 | +474% | 0 | 0 | — |
case-06 | fail→pass | 4,845 | 16,466 | +240% | 1 | 1 | 0% | 869 | 3,167 | +264% | 0 | 0 | — |
case-07 | fail→pass | 12,830 | 24,098 | +88% | 1 | 1 | 0% | 2,041 | 3,111 | +52% | 0 | 0 | — |
case-08 | fail→pass | 8,858 | 28,285 | +219% | 1 | 1 | 0% | 1,290 | 5,694 | +341% | 0 | 0 | — |
case-09 | fail→fail | 25,477 | 13,211 | -48% | 1 | 1 | 0% | 4,378 | 2,352 | -46% | 0 | 0 | — |
case-10 | fail→pass | 4,174 | 13,095 | +214% | 1 | 1 | 0% | 750 | 2,537 | +238% | 0 | 0 | — |
case-11 | fail→pass | 3,203 | 13,152 | +311% | 1 | 1 | 0% | 370 | 2,723 | +636% | 0 | 0 | — |
case-12 | pass→fail | 13,646 | 10,591 | -22% | 1 | 1 | 0% | 2,333 | 2,075 | -11% | 0 | 0 | — |
case-13 | fail→fail | 3,466 | 23,963 | +591% | 1 | 1 | 0% | 517 | 4,579 | +786% | 0 | 0 | — |
case-14 | fail→pass | 4,446 | 9,523 | +114% | 1 | 1 | 0% | 779 | 2,158 | +177% | 0 | 0 | — |
case-15 | fail→pass | 2,084 | 8,923 | +328% | 1 | 1 | 0% | 305 | 1,803 | +491% | 0 | 0 | — |
case-16 | fail→pass | 14,364 | 16,492 | +15% | 1 | 1 | 0% | 2,751 | 3,436 | +25% | 0 | 0 | — |
case-17 | fail→pass | 9,305 | 9,772 | +5% | 1 | 1 | 0% | 1,369 | 1,884 | +38% | 0 | 0 | — |
case-18 | pass→fail | 9,301 | 18,017 | +94% | 1 | 1 | 0% | 1,581 | 3,525 | +123% | 0 | 0 | — |
case-19 | fail→fail | 8,745 | 12,237 | +40% | 1 | 1 | 0% | 1,500 | 2,510 | +67% | 0 | 0 | — |
case-20 | pass→fail | 18,347 | 27,179 | +48% | 1 | 1 | 0% | 4,482 | 6,420 | +43% | 0 | 0 | — |
case-21 | fail→fail | 2,101 | 11,066 | +427% | 1 | 1 | 0% | 341 | 2,416 | +609% | 0 | 0 | — |
case-22 | pass→pass | 9,974 | 8,168 | -18% | 1 | 1 | 0% | 2,047 | 1,948 | -5% | 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 +41 percentage points is the difference between those two pass rates over the 22 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.