Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →识别账户余额异常波动、大额资金进出、AUM剧烈变化和异常活跃，辅助重点账户监测与风险预警
.claude/skills/aifinlab-account-volatility-monitoring/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 13% | 0% |
银行需要持续监测账户层面的异常波动，包括余额大幅变动、大额资金集中进出、AUM短期剧变、长期沉寂后突然激活等情况。这些波动可能反映客户资金搬家、流失风险、异常交易行为或潜在风控信号，需要及时识别并分级关注。
核心问题包括：
| 字段 | 说明 | 是否必需 | |------|------|----------| | 监测时段 | 日/周/月/自定义区间 | 是 | | 监测对象 | 账户/客户/客群/机构 | 是 | | 账户余额数据 | 含日终余额、AUM等时序数据 | 是 | | 交易流水数据 | 含金额、方向、时间戳等 | 否 | | 波动阈值 | 自定义波动触发标准 | 否 | | 历史基线 | 账户历史余额/活跃度基线 | 否 |
bank-calc-utils：提供余额变动计算、波动率统计、阈值判定等基础计算能力large-fund-movement-tracking：对波动账户中涉及大额资金变动的进行深入追踪python# 调用 skill result = run_skill({ "param1": "value1", "param2": "value2" })
bashpython scripts/run_skill.py --input data.json
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 34,669 | 42,687 | +23% | 1 | 1 | 0% | 5,313 | 5,023 | -5% | 0 | 0 | — |
case-02 | fail→fail | 27,596 | 31,889 | +16% | 1 | 1 | 0% | 4,139 | 5,845 | +41% | 0 | 0 | — |
case-03 | pass→pass | 14,175 | 9,776 | -31% | 1 | 1 | 0% | 1,997 | 2,612 | +31% | 0 | 0 | — |
case-04 | pass→pass | 10,418 | 9,529 | -9% | 1 | 1 | 0% | 1,697 | 2,515 | +48% | 0 | 0 | — |
case-05 | fail→pass | 16,569 | 14,683 | -11% | 1 | 1 | 0% | 2,320 | 2,895 | +25% | 0 | 0 | — |
case-06 | pass→pass | 19,276 | 16,473 | -15% | 1 | 1 | 0% | 2,541 | 3,509 | +38% | 0 | 0 | — |
case-07 | fail→pass | 14,502 | 5,033 | -65% | 1 | 1 | 0% | 2,008 | 1,795 | -11% | 0 | 0 | — |
case-08 | fail→pass | 14,544 | 3,120 | -79% | 1 | 1 | 0% | 2,270 | 1,443 | -36% | 0 | 0 | — |
case-09 | pass→pass | 16,757 | 8,452 | -50% | 1 | 1 | 0% | 2,354 | 2,275 | -3% | 0 | 0 | — |
case-10 | pass→pass | 20,612 | 15,556 | -25% | 1 | 1 | 0% | 2,816 | 3,483 | +24% | 0 | 0 | — |
case-11 | pass→pass | 15,938 | 7,073 | -56% | 1 | 1 | 0% | 2,352 | 2,092 | -11% | 0 | 0 | — |
case-12 | pass→pass | 16,489 | 12,802 | -22% | 1 | 1 | 0% | 2,574 | 3,058 | +19% | 0 | 0 | — |
case-13 | pass→pass | 17,887 | 17,853 | -0% | 1 | 1 | 0% | 2,467 | 3,313 | +34% | 0 | 0 | — |
case-14 | pass→pass | 21,199 | 22,099 | +4% | 1 | 1 | 0% | 2,589 | 3,839 | +48% | 0 | 0 | — |
case-15 | pass→pass | 11,066 | 6,739 | -39% | 1 | 1 | 0% | 1,322 | 2,113 | +60% | 0 | 0 | — |
case-16 | fail→pass | 23,215 | 12,558 | -46% | 1 | 1 | 0% | 2,813 | 2,637 | -6% | 0 | 0 | — |
case-17 | pass→pass | 12,435 | 4,882 | -61% | 1 | 1 | 0% | 2,218 | 1,568 | -29% | 0 | 0 | — |
case-18 | pass→pass | 12,494 | 12,260 | -2% | 1 | 1 | 0% | 1,848 | 2,869 | +55% | 0 | 0 | — |
case-19 | pass→pass | 10,248 | 6,247 | -39% | 1 | 1 | 0% | 1,506 | 2,121 | +41% | 0 | 0 | — |
case-20 | fail→pass | 24,371 | 20,282 | -17% | 1 | 1 | 0% | 3,074 | 3,469 | +13% | 0 | 0 | — |
case-21 | fail→fail | 17,185 | 22,066 | +28% | 1 | 1 | 0% | 2,740 | 3,232 | +18% | 0 | 0 | — |
case-22 | fail→pass | 12,241 | 10,163 | -17% | 1 | 1 | 0% | 1,881 | 2,543 | +35% | 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 +27 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.