Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Incorporate a new judgment into the rating model and return updated ratings for all candidates.
.claude/skills/yogsoth-ai-rating-update/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | -7% | 0% |
| case-10 | ✓→✓ | ＝ Same ✓ | -15% | 0% |
| case-18 | ✓→✓ | ＝ Same ✓ | 27% | 0% |
Incorporates a new pairwise judgment into the current rating model. Applies the specified method's update rules (Bradley-Terry MLE, Elo K-factor, TrueSkill Gaussian update, etc.) and returns the full updated rating vector.
Runs as a subagent. Receives judgment, current ratings, and method specification, returns updated ratings.
Rating updates involve mathematical computation specific to the chosen method. Isolating this ensures correct application of update formulas without mixing method-specific logic into orchestration.
Output MUST contain updated ratings for ALL candidates (not just the compared pair). The winner's rating MUST increase or stay the same. The method field MUST match the input method.
<!-- 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→fail | 10,343 | 11,757 | +14% | 1 | 1 | 0% | 1,252 | 1,699 | +36% | 0 | 0 | — |
case-02 | fail→fail | 21,492 | 22,140 | +3% | 1 | 1 | 0% | 3,447 | 5,133 | +49% | 0 | 0 | — |
case-03 | fail→fail | 39,134 | 20,366 | -48% | 1 | 1 | 0% | 6,302 | 3,211 | -49% | 0 | 0 | — |
case-04 | fail→fail | 13,043 | 10,640 | -18% | 1 | 1 | 0% | 1,560 | 1,397 | -10% | 0 | 0 | — |
case-05 | fail→fail | 30,244 | 38,467 | +27% | 1 | 1 | 0% | 5,044 | 7,215 | +43% | 0 | 0 | — |
case-06 | fail→fail | 13,618 | 26,820 | +97% | 1 | 1 | 0% | 3,006 | 4,969 | +65% | 0 | 0 | — |
case-07 | pass→pass | 14,770 | 13,311 | -10% | 1 | 1 | 0% | 2,144 | 1,985 | -7% | 0 | 0 | — |
case-08 | fail→fail | 17,480 | 18,248 | +4% | 1 | 1 | 0% | 2,743 | 4,245 | +55% | 0 | 0 | — |
case-09 | fail→fail | 10,210 | 13,384 | +31% | 1 | 1 | 0% | 2,356 | 2,412 | +2% | 0 | 0 | — |
case-15 | fail→pass | 17,918 | 21,967 | +23% | 1 | 1 | 0% | 2,615 | 3,869 | +48% | 0 | 0 | — |
case-10 | pass→pass | 14,663 | 12,557 | -14% | 1 | 1 | 0% | 2,235 | 1,902 | -15% | 0 | 0 | — |
case-11 | fail→fail | 22,789 | 34,080 | +50% | 1 | 1 | 0% | 3,841 | 6,625 | +72% | 0 | 0 | — |
case-12 | fail→fail | 13,652 | 12,700 | -7% | 1 | 1 | 0% | 1,737 | 1,838 | +6% | 0 | 0 | — |
case-13 | fail→pass | 21,183 | 12,521 | -41% | 1 | 1 | 0% | 3,372 | 2,680 | -21% | 0 | 0 | — |
case-14 | fail→fail | 12,143 | 4,953 | -59% | 1 | 1 | 0% | 1,515 | 1,114 | -26% | 0 | 0 | — |
case-16 | fail→fail | 9,824 | 11,629 | +18% | 1 | 1 | 0% | 1,492 | 1,687 | +13% | 0 | 0 | — |
case-17 | fail→fail | 16,517 | 22,374 | +35% | 1 | 1 | 0% | 3,755 | 5,788 | +54% | 0 | 0 | — |
case-18 | pass→pass | 14,425 | 14,311 | -1% | 1 | 1 | 0% | 1,931 | 2,455 | +27% | 0 | 0 | — |
case-19 | fail→fail | 17,025 | 14,319 | -16% | 1 | 1 | 0% | 2,128 | 3,105 | +46% | 0 | 0 | — |
case-20 | pass→pass | 5,185 | 7,362 | +42% | 1 | 1 | 0% | 1,112 | 1,628 | +46% | 0 | 0 | — |
case-21 | pass→pass | 6,991 | 6,246 | -11% | 1 | 1 | 0% | 827 | 1,006 | +22% | 0 | 0 | — |
case-22 | pass→pass | 7,156 | 3,554 | -50% | 1 | 1 | 0% | 539 | 714 | +32% | 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 +9 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.