Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要在银行对公金融场景下，围绕流水核验进行完整性、一致性、真实性或规则符合性检查时使用本技能。适合输出核验结论、异常项清单、补件要求和升级复核建议。
.claude/skills/aifinlab-bank-t148-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 23% | 0% |
本技能用于对对公交易流水做“可复核”的核验输出：先查样本覆盖与字段完整性，再查一致性（账户/对手/方向/余额/用途），最后识别异常特征并给出补件与升级复核建议。输出强调证据链与可追溯性。
company.name + verification.time_window + verification.transactionsreferences/input-schema.mdbashpython scripts/run_skill.py --input assets/example-input.json --format markdown python scripts/run_skill.py --input assets/example-input.json --format json
脚本入口：scripts/run_skill.py（调用 shared/corporate_ops_skill_engine.py 的 t148 场景）。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-08 | pass→pass | 16,259 | 22,769 | +40% | 1 | 1 | 0% | 2,074 | 3,962 | +91% | 0 | 0 | — |
case-09 | fail→pass | 11,142 | 4,161 | -63% | 1 | 1 | 0% | 1,853 | 1,361 | -27% | 0 | 0 | — |
case-10 | fail→pass | 13,892 | 3,112 | -78% | 1 | 1 | 0% | 2,056 | 1,282 | -38% | 0 | 0 | — |
case-07 | pass→pass | 26,501 | 51,018 | +93% | 1 | 1 | 0% | 3,875 | 4,805 | +24% | 0 | 0 | — |
case-01 | fail→fail | 29,419 | 62,475 | +112% | 1 | 1 | 0% | 3,674 | 5,834 | +59% | 0 | 0 | — |
case-02 | fail→pass | 22,180 | 30,981 | +40% | 1 | 1 | 0% | 3,192 | 5,249 | +64% | 0 | 0 | — |
case-03 | pass→pass | 26,458 | 25,641 | -3% | 1 | 1 | 0% | 3,412 | 4,761 | +40% | 0 | 0 | — |
case-04 | pass→pass | 18,822 | 25,948 | +38% | 1 | 1 | 0% | 2,670 | 4,025 | +51% | 0 | 0 | — |
case-05 | pass→pass | 20,641 | 18,615 | -10% | 1 | 1 | 0% | 2,423 | 3,602 | +49% | 0 | 0 | — |
case-06 | pass→pass | 23,059 | 25,067 | +9% | 1 | 1 | 0% | 3,347 | 4,569 | +37% | 0 | 0 | — |
case-11 | pass→pass | 13,924 | 14,687 | +5% | 1 | 1 | 0% | 1,923 | 2,152 | +12% | 0 | 0 | — |
case-12 | pass→pass | 21,129 | 29,898 | +42% | 1 | 1 | 0% | 3,107 | 3,741 | +20% | 0 | 0 | — |
case-13 | pass→pass | 23,935 | 22,527 | -6% | 1 | 1 | 0% | 2,950 | 4,251 | +44% | 0 | 0 | — |
case-14 | pass→pass | 20,211 | 29,602 | +46% | 1 | 1 | 0% | 2,772 | 4,541 | +64% | 0 | 0 | — |
case-15 | pass→pass | 22,713 | 22,553 | -1% | 1 | 1 | 0% | 2,874 | 4,055 | +41% | 0 | 0 | — |
case-16 | pass→pass | 17,637 | 22,017 | +25% | 1 | 1 | 0% | 2,358 | 3,592 | +52% | 0 | 0 | — |
case-17 | pass→pass | 18,198 | 21,629 | +19% | 1 | 1 | 0% | 2,761 | 3,629 | +31% | 0 | 0 | — |
case-18 | pass→pass | 20,553 | 21,016 | +2% | 1 | 1 | 0% | 2,568 | 3,353 | +31% | 0 | 0 | — |
case-19 | fail→pass | 22,021 | 20,983 | -5% | 1 | 1 | 0% | 2,976 | 3,969 | +33% | 0 | 0 | — |
case-20 | fail→pass | 14,449 | 12,263 | -15% | 1 | 1 | 0% | 2,156 | 2,661 | +23% | 0 | 0 | — |
case-21 | pass→pass | 15,555 | 16,936 | +9% | 1 | 1 | 0% | 2,182 | 3,272 | +50% | 0 | 0 | — |
case-22 | fail→fail | 16,384 | 14,141 | -14% | 1 | 1 | 0% | 1,100 | 2,531 | +130% | 0 | 0 | — |
case-23 | fail→fail | 15,980 | 14,920 | -7% | 1 | 1 | 0% | 2,874 | 2,872 | -0% | 0 | 0 | — |
case-24 | fail→pass | 23,549 | 17,416 | -26% | 1 | 1 | 0% | 2,932 | 2,943 | +0% | 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. 24 cases were attempted. The headline lift of +25 percentage points is the difference between those two pass rates over the 24 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.