Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要在银行风险管理场景下，对贷中审批校验进行初筛、尽调支撑或准入前梳理时使用本技能。适合输出面向风控与预警团队的结构化判断、待补资料清单、关键风险提示和下一步推进建议，并可配合规则校验脚本完成一致性输出。
.claude/skills/aifinlab-bank-t197-risk-management-approval-validation-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 19% | 0% |
本技能面向银行风险管理与预警运营场景，目标是把规则命中、趋势变化和处置建议整合成可跟进的风险视图，而不是只给指标或告警日志。输出结果需要支持风控团队快速判断“是否能推进”“要补什么”“要升级到哪一步”。
当用户提供规则命中清单与样本指标时，可调用脚本生成结构化校验结果，保证输出格式统一。
scripts/approval_rule_validator.pyrules.json（规则定义）、samples.csv（样本指标）validation_report.json（命中汇总、优先级、待核验清单）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 23,963 | 20,547 | -14% | 1 | 1 | 0% | 3,122 | 3,906 | +25% | 0 | 0 | — |
case-02 | pass→pass | 27,738 | 22,972 | -17% | 1 | 1 | 0% | 3,474 | 4,445 | +28% | 0 | 0 | — |
case-03 | fail→pass | 28,386 | 25,308 | -11% | 1 | 1 | 0% | 3,551 | 4,156 | +17% | 0 | 0 | — |
case-04 | pass→pass | 14,106 | 18,800 | +33% | 1 | 1 | 0% | 2,092 | 3,370 | +61% | 0 | 0 | — |
case-05 | pass→pass | 12,397 | 14,476 | +17% | 1 | 1 | 0% | 1,636 | 3,009 | +84% | 0 | 0 | — |
case-06 | fail→pass | 22,147 | 19,271 | -13% | 1 | 1 | 0% | 2,792 | 3,616 | +30% | 0 | 0 | — |
case-07 | fail→fail | 27,125 | 22,119 | -18% | 1 | 1 | 0% | 3,237 | 3,798 | +17% | 0 | 0 | — |
case-08 | fail→pass | 27,433 | 21,326 | -22% | 1 | 1 | 0% | 3,545 | 4,107 | +16% | 0 | 0 | — |
case-09 | fail→pass | 25,130 | 21,280 | -15% | 1 | 1 | 0% | 3,030 | 3,596 | +19% | 0 | 0 | — |
case-10 | fail→pass | 23,350 | 14,394 | -38% | 1 | 1 | 0% | 3,224 | 3,498 | +8% | 0 | 0 | — |
case-11 | fail→pass | 18,637 | 17,270 | -7% | 1 | 1 | 0% | 2,648 | 3,374 | +27% | 0 | 0 | — |
case-12 | fail→fail | 20,573 | 20,848 | +1% | 1 | 1 | 0% | 3,138 | 3,993 | +27% | 0 | 0 | — |
case-13 | pass→pass | 22,910 | 26,180 | +14% | 1 | 1 | 0% | 3,352 | 4,199 | +25% | 0 | 0 | — |
case-14 | fail→pass | 19,936 | 20,738 | +4% | 1 | 1 | 0% | 2,491 | 3,540 | +42% | 0 | 0 | — |
case-15 | fail→fail | 21,242 | 24,423 | +15% | 1 | 1 | 0% | 3,119 | 4,008 | +29% | 0 | 0 | — |
case-16 | fail→pass | 21,806 | 19,256 | -12% | 1 | 1 | 0% | 3,045 | 3,460 | +14% | 0 | 0 | — |
case-17 | fail→fail | 20,697 | 21,485 | +4% | 1 | 1 | 0% | 2,966 | 3,624 | +22% | 0 | 0 | — |
case-18 | fail→pass | 28,011 | 22,707 | -19% | 1 | 1 | 0% | 3,252 | 3,980 | +22% | 0 | 0 | — |
case-19 | fail→pass | 17,992 | 21,292 | +18% | 1 | 1 | 0% | 2,490 | 3,976 | +60% | 0 | 0 | — |
case-20 | fail→pass | 24,278 | 21,754 | -10% | 1 | 1 | 0% | 2,736 | 3,909 | +43% | 0 | 0 | — |
case-21 | fail→pass | 25,065 | 24,987 | -0% | 1 | 1 | 0% | 3,146 | 3,649 | +16% | 0 | 0 | — |
case-22 | fail→fail | 18,755 | 18,950 | +1% | 1 | 1 | 0% | 2,729 | 3,399 | +25% | 0 | 0 | — |
case-23 | fail→pass | 17,891 | 24,135 | +35% | 1 | 1 | 0% | 2,781 | 3,926 | +41% | 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 +61 percentage points is the difference between those two pass rates over the 23 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.