Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when identifying retail credit-card limit increase opportunities or prioritizing outreach; trigger for requests that need Chinese skill content with clear eligibility checks, ranking logic, and optional scripts for scoring.
.claude/skills/aifinlab-bank-t157-retail-finance-limit-increase-opportunity-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 82% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 14% | 0% |
用于识别信用卡提额机会、排序潜力客户并输出触达建议与风控边界。它只做“机会识别与优先级建议”，不替代授信审批与最终额度决策。
customer_id、现有额度、卡片类型批量评分使用 scripts/limit_increase_scoring.py。
bashpython scripts/limit_increase_scoring.py --input limit_candidates.json --output limit_scored.json
输入 JSON 示例：
json{ "customers": [ { "customer_id": "C002", "credit_limit": 20000, "avg_spend_3m": 16000, "on_time_rate_12m": 0.99, "delinquency_12m": 0, "monthly_income": 15000, "debt_to_income": 0.35, "risk_flag": 0 } ] }
输出关键字段：
decision：建议提额/可观察/暂缓score：规则评分（用于排序）reasons：主要触发因素text客户：{customer_id} 评分：{score} 建议：{decision} 理由：{reasons} 待补信息：{missing_fields}
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 27,195 | 19,983 | -27% | 1 | 1 | 0% | 3,635 | 3,873 | +7% | 0 | 0 | — |
case-02 | fail→pass | 14,532 | 14,773 | +2% | 1 | 1 | 0% | 2,364 | 3,148 | +33% | 0 | 0 | — |
case-03 | fail→fail | 23,901 | 25,556 | +7% | 1 | 1 | 0% | 3,424 | 4,003 | +17% | 0 | 0 | — |
case-04 | fail→pass | 11,170 | 13,677 | +22% | 1 | 1 | 0% | 1,481 | 2,694 | +82% | 0 | 0 | — |
case-05 | pass→pass | 16,453 | 12,810 | -22% | 1 | 1 | 0% | 2,311 | 2,346 | +2% | 0 | 0 | — |
case-06 | pass→pass | 13,693 | 9,750 | -29% | 1 | 1 | 0% | 1,934 | 1,911 | -1% | 0 | 0 | — |
case-07 | fail→pass | 20,240 | 6,805 | -66% | 1 | 1 | 0% | 2,624 | 2,096 | -20% | 0 | 0 | — |
case-08 | fail→pass | 15,333 | 11,273 | -26% | 1 | 1 | 0% | 2,315 | 2,510 | +8% | 0 | 0 | — |
case-09 | fail→pass | 18,158 | 12,730 | -30% | 1 | 1 | 0% | 2,338 | 2,672 | +14% | 0 | 0 | — |
case-10 | fail→pass | 19,453 | 17,081 | -12% | 1 | 1 | 0% | 2,727 | 3,278 | +20% | 0 | 0 | — |
case-11 | pass→pass | 22,600 | 17,677 | -22% | 1 | 1 | 0% | 2,874 | 2,901 | +1% | 0 | 0 | — |
case-12 | fail→pass | 23,299 | 20,479 | -12% | 1 | 1 | 0% | 3,327 | 3,460 | +4% | 0 | 0 | — |
case-13 | fail→pass | 33,575 | 7,553 | -78% | 1 | 1 | 0% | 2,639 | 1,725 | -35% | 0 | 0 | — |
case-14 | pass→pass | 14,134 | 11,875 | -16% | 1 | 1 | 0% | 2,724 | 2,332 | -14% | 0 | 0 | — |
case-15 | pass→pass | 15,388 | 17,022 | +11% | 1 | 1 | 0% | 2,258 | 2,965 | +31% | 0 | 0 | — |
case-16 | pass→pass | 9,397 | 2,809 | -70% | 1 | 1 | 0% | 1,358 | 1,246 | -8% | 0 | 0 | — |
case-17 | pass→pass | 29,135 | 15,377 | -47% | 1 | 1 | 0% | 2,356 | 3,069 | +30% | 0 | 0 | — |
case-18 | pass→pass | 19,803 | 16,426 | -17% | 1 | 1 | 0% | 2,803 | 2,974 | +6% | 0 | 0 | — |
case-19 | pass→pass | 16,176 | 9,949 | -38% | 1 | 1 | 0% | 2,424 | 2,572 | +6% | 0 | 0 | — |
case-20 | fail→pass | 17,088 | 9,279 | -46% | 1 | 1 | 0% | 2,156 | 2,086 | -3% | 0 | 0 | — |
case-21 | fail→pass | 19,976 | 19,486 | -2% | 1 | 1 | 0% | 2,747 | 3,302 | +20% | 0 | 0 | — |
case-22 | fail→pass | 16,516 | 12,800 | -22% | 1 | 1 | 0% | 2,058 | 2,837 | +38% | 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 +50 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.