Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当需要检索业务SOP并输出可执行步骤、责任归属和合规提醒时使用本技能，适用于运营、合规与内控人员快速定位操作流程。
.claude/skills/aifinlab-bank-t225-compliance-operations-sop-retrieval-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 52% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 1% | 0% |
本技能用于根据问题与业务场景检索SOP条款，整理为可执行步骤与责任分工，强调口径一致、可追溯与可留痕。
scripts/sop_retrieval_builder.py| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 34,351 | 21,470 | -37% | 1 | 1 | 0% | 4,586 | 3,490 | -24% | 0 | 0 | — |
case-02 | fail→fail | 27,021 | 27,213 | +1% | 1 | 1 | 0% | 3,847 | 3,987 | +4% | 0 | 0 | — |
case-03 | fail→fail | 28,501 | 25,312 | -11% | 1 | 1 | 0% | 4,112 | 3,644 | -11% | 0 | 0 | — |
case-04 | pass→pass | 27,976 | 34,772 | +24% | 1 | 1 | 0% | 3,399 | 4,464 | +31% | 0 | 0 | — |
case-05 | fail→pass | 16,934 | 26,130 | +54% | 1 | 1 | 0% | 2,542 | 3,873 | +52% | 0 | 0 | — |
case-06 | pass→pass | 13,038 | 19,463 | +49% | 1 | 1 | 0% | 1,551 | 2,259 | +46% | 0 | 0 | — |
case-07 | pass→pass | 24,700 | 21,942 | -11% | 1 | 1 | 0% | 3,219 | 3,286 | +2% | 0 | 0 | — |
case-08 | fail→pass | 26,808 | 26,877 | +0% | 1 | 1 | 0% | 3,488 | 3,944 | +13% | 0 | 0 | — |
case-09 | fail→pass | 30,474 | 18,231 | -40% | 1 | 1 | 0% | 4,267 | 2,971 | -30% | 0 | 0 | — |
case-10 | pass→pass | 32,159 | 25,470 | -21% | 1 | 1 | 0% | 4,109 | 3,520 | -14% | 0 | 0 | — |
case-11 | fail→pass | 22,295 | 15,700 | -30% | 1 | 1 | 0% | 2,704 | 2,778 | +3% | 0 | 0 | — |
case-12 | pass→pass | 25,208 | 29,633 | +18% | 1 | 1 | 0% | 3,642 | 4,120 | +13% | 0 | 0 | — |
case-13 | fail→pass | 33,150 | 25,404 | -23% | 1 | 1 | 0% | 4,312 | 4,352 | +1% | 0 | 0 | — |
case-14 | pass→fail | 38,399 | 29,692 | -23% | 1 | 1 | 0% | 4,881 | 4,690 | -4% | 0 | 0 | — |
case-15 | fail→pass | 18,019 | 21,568 | +20% | 1 | 1 | 0% | 1,851 | 3,171 | +71% | 0 | 0 | — |
case-16 | pass→pass | 9,450 | 15,234 | +61% | 1 | 1 | 0% | 1,443 | 2,458 | +70% | 0 | 0 | — |
case-17 | fail→fail | 35,911 | 27,994 | -22% | 1 | 1 | 0% | 3,721 | 3,907 | +5% | 0 | 0 | — |
case-18 | fail→pass | 22,546 | 19,761 | -12% | 1 | 1 | 0% | 2,889 | 3,248 | +12% | 0 | 0 | — |
case-19 | fail→pass | 22,335 | 23,351 | +5% | 1 | 1 | 0% | 3,342 | 3,955 | +18% | 0 | 0 | — |
case-20 | pass→pass | 30,656 | 19,151 | -38% | 1 | 1 | 0% | 4,285 | 3,765 | -12% | 0 | 0 | — |
case-21 | fail→pass | 12,257 | 31,620 | +158% | 1 | 1 | 0% | 1,941 | 2,829 | +46% | 0 | 0 | — |
case-22 | fail→pass | 24,031 | 32,837 | +37% | 1 | 1 | 0% | 3,471 | 4,079 | +18% | 0 | 0 | — |
case-23 | pass→pass | 22,522 | 23,253 | +3% | 1 | 1 | 0% | 3,603 | 4,010 | +11% | 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. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 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.