Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Evaluate whether the ranking has stabilized by analyzing rating history and computing stability metrics.
.claude/skills/yogsoth-ai-convergence-check/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 160% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 36% | 0% |
Evaluates whether the current ranking has converged by analyzing the trajectory of ratings over recent iterations. Computes stability metrics and determines if further comparisons would meaningfully change the ranking.
Runs as a subagent. Receives the full rating history and returns convergence status with metrics.
Convergence assessment requires analyzing trends across multiple snapshots and applying statistical tests. Isolating this prevents the orchestrator from needing to hold the full history in working memory.
Output MUST contain a boolean converged field and a numeric stability_score in 0, 1]. If converged=false, MUST include recommendation for what to do next.
<!-- 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 | 4,751 | 10,353 | +118% | 1 | 1 | 0% | 733 | 1,909 | +160% | 0 | 0 | — |
case-02 | fail→pass | 8,448 | 11,437 | +35% | 1 | 1 | 0% | 1,278 | 2,009 | +57% | 0 | 0 | — |
case-03 | fail→pass | 8,068 | 8,375 | +4% | 1 | 1 | 0% | 1,323 | 1,586 | +20% | 0 | 0 | — |
case-04 | fail→pass | 9,065 | 8,035 | -11% | 1 | 1 | 0% | 1,451 | 1,572 | +8% | 0 | 0 | — |
case-05 | fail→pass | 9,226 | 10,673 | +16% | 1 | 1 | 0% | 1,517 | 2,064 | +36% | 0 | 0 | — |
case-06 | fail→pass | 13,285 | 7,191 | -46% | 1 | 1 | 0% | 1,929 | 1,341 | -30% | 0 | 0 | — |
case-07 | fail→pass | 10,601 | 4,474 | -58% | 1 | 1 | 0% | 1,608 | 932 | -42% | 0 | 0 | — |
case-08 | fail→pass | 16,320 | 7,459 | -54% | 1 | 1 | 0% | 2,466 | 1,411 | -43% | 0 | 0 | — |
case-09 | pass→pass | 11,874 | 3,819 | -68% | 1 | 1 | 0% | 1,857 | 840 | -55% | 0 | 0 | — |
case-10 | pass→pass | 10,600 | 3,969 | -63% | 1 | 1 | 0% | 1,657 | 836 | -50% | 0 | 0 | — |
case-11 | fail→pass | 6,957 | 8,505 | +22% | 1 | 1 | 0% | 1,103 | 1,695 | +54% | 0 | 0 | — |
case-12 | pass→pass | 13,561 | 6,096 | -55% | 1 | 1 | 0% | 2,060 | 1,044 | -49% | 0 | 0 | — |
case-13 | pass→pass | 10,967 | 5,928 | -46% | 1 | 1 | 0% | 1,664 | 1,162 | -30% | 0 | 0 | — |
case-14 | fail→fail | 11,067 | 1,883 | -83% | 1 | 1 | 0% | 1,641 | 540 | -67% | 0 | 0 | — |
case-15 | fail→fail | 3,231 | 9,368 | +190% | 1 | 1 | 0% | 495 | 1,797 | +263% | 0 | 0 | — |
case-16 | pass→pass | 12,403 | 11,878 | -4% | 1 | 1 | 0% | 2,568 | 2,617 | +2% | 0 | 0 | — |
case-17 | fail→pass | 9,912 | 16,960 | +71% | 1 | 1 | 0% | 1,672 | 3,291 | +97% | 0 | 0 | — |
case-18 | pass→pass | 9,626 | 9,736 | +1% | 1 | 1 | 0% | 1,601 | 1,750 | +9% | 0 | 0 | — |
case-19 | pass→pass | 16,987 | 10,506 | -38% | 1 | 1 | 0% | 2,497 | 1,792 | -28% | 0 | 0 | — |
case-20 | fail→pass | 10,654 | 3,284 | -69% | 1 | 1 | 0% | 1,646 | 710 | -57% | 0 | 0 | — |
case-21 | fail→pass | 5,323 | 2,857 | -46% | 1 | 1 | 0% | 988 | 657 | -34% | 0 | 0 | — |
case-22 | fail→pass | 10,937 | 5,908 | -46% | 1 | 1 | 0% | 1,646 | 1,053 | -36% | 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 +59 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.