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，对逾期苗头进行持续监测、规则命中解释、风险分层和处置建议输出时使用本技能，适合生成风险摘要、优先级判断与升级路径。
.claude/skills/aifinlab-bank-t199-risk-management-delinquency-identification-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-20 | ✓→✗ | ▼ Worse | 69% | 0% |
本技能用于银行风险管理的逾期苗头识别与风险分层，目标是把规则命中、趋势变化、客户/账户行为特征转化为可执行的排查与处置建议，输出清晰的优先级和升级路径，而不是只罗列指标或告警。
当需要批量计算逾期苗头信号与优先级分层时，使用 scripts/delinquency_signal_builder.py：
输入 CSV 需包含字段（可按需要增减）：
customer_id 客户号account_id 账户号product_type 产品类型statement_date 账期日期due_amount 应还金额paid_amount 实还金额days_past_due 逾期天数rule_hits 命中规则数量balance 余额historical_avg_dpd 历史平均逾期天数输出字段：
risk_level 风险等级（High/Medium/Low）risk_score 风险评分（0-100）signal_summary 信号摘要next_action 推荐动作missing_fields 缺失关键字段| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 23,130 | 24,975 | +8% | 1 | 1 | 0% | 2,527 | 3,425 | +36% | 0 | 0 | — |
case-17 | pass→pass | 19,090 | 24,075 | +26% | 1 | 1 | 0% | 2,406 | 4,293 | +78% | 0 | 0 | — |
case-18 | pass→pass | 15,248 | 17,763 | +16% | 1 | 1 | 0% | 2,319 | 3,886 | +68% | 0 | 0 | — |
case-02 | pass→pass | 19,164 | 16,962 | -11% | 1 | 1 | 0% | 1,784 | 3,435 | +93% | 0 | 0 | — |
case-03 | pass→pass | 17,694 | 19,307 | +9% | 1 | 1 | 0% | 1,822 | 3,921 | +115% | 0 | 0 | — |
case-04 | fail→pass | 11,114 | 8,600 | -23% | 1 | 1 | 0% | 2,231 | 2,767 | +24% | 0 | 0 | — |
case-05 | fail→pass | 15,339 | 4,539 | -70% | 1 | 1 | 0% | 2,358 | 1,664 | -29% | 0 | 0 | — |
case-19 | pass→pass | 17,716 | 19,431 | +10% | 1 | 1 | 0% | 2,332 | 3,632 | +56% | 0 | 0 | — |
case-06 | pass→pass | 10,067 | 2,369 | -76% | 1 | 1 | 0% | 1,342 | 1,387 | +3% | 0 | 0 | — |
case-07 | pass→pass | 10,696 | 17,619 | +65% | 1 | 1 | 0% | 1,796 | 3,401 | +89% | 0 | 0 | — |
case-08 | pass→pass | 17,706 | 20,548 | +16% | 1 | 1 | 0% | 2,678 | 3,880 | +45% | 0 | 0 | — |
case-09 | pass→pass | 12,496 | 15,888 | +27% | 1 | 1 | 0% | 2,080 | 3,119 | +50% | 0 | 0 | — |
case-10 | pass→pass | 18,535 | 24,994 | +35% | 1 | 1 | 0% | 2,787 | 4,263 | +53% | 0 | 0 | — |
case-11 | pass→pass | 12,367 | 15,524 | +26% | 1 | 1 | 0% | 1,625 | 3,116 | +92% | 0 | 0 | — |
case-12 | pass→pass | 13,817 | 17,121 | +24% | 1 | 1 | 0% | 1,869 | 3,277 | +75% | 0 | 0 | — |
case-13 | pass→pass | 17,032 | 15,916 | -7% | 1 | 1 | 0% | 2,476 | 3,388 | +37% | 0 | 0 | — |
case-14 | pass→pass | 20,222 | 20,603 | +2% | 1 | 1 | 0% | 2,603 | 4,318 | +66% | 0 | 0 | — |
case-15 | fail→pass | 12,526 | 8,946 | -29% | 1 | 1 | 0% | 1,587 | 2,518 | +59% | 0 | 0 | — |
case-16 | pass→pass | 8,744 | 16,993 | +94% | 1 | 1 | 0% | 1,399 | 3,685 | +163% | 0 | 0 | — |
case-20 | pass→fail | 16,297 | 20,157 | +24% | 1 | 1 | 0% | 2,490 | 4,212 | +69% | 0 | 0 | — |
case-21 | pass→pass | 20,381 | 21,112 | +4% | 1 | 1 | 0% | 2,473 | 3,893 | +57% | 0 | 0 | — |
case-22 | pass→pass | 12,042 | 16,082 | +34% | 1 | 1 | 0% | 1,646 | 3,259 | +98% | 0 | 0 | — |
case-23 | pass→pass | 13,011 | 14,311 | +10% | 1 | 1 | 0% | 1,773 | 3,482 | +96% | 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 +13 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.