Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when performing retail mortgage pre-checks or document gap analysis before formal approval; trigger for requests that need Chinese skill content with structured precheck logic, outputs, and optional scripts.
.claude/skills/aifinlab-bank-t159-retail-finance-pre-check-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 56% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -8% | 0% |
面向房贷业务的预审初筛与资料梳理，输出“可推进/待补资料/阻断”级别的初步判断、关键风险提示与补件清单，服务客户经理与运营团队的前置沟通。
applicant_id、年龄、职业、婚姻/家庭情况可推进/待补资料/阻断批量预审使用 scripts/precheck_mortgage.py。
bashpython scripts/precheck_mortgage.py --input mortgage_applications.json --output mortgage_precheck.json
输入 JSON 示例：
json{ "applications": [ { "applicant_id": "A001", "age": 35, "credit_score": 680, "debt_to_income": 0.42, "down_payment_ratio": 0.35, "property_type": "住宅", "income_verified": true } ] }
输出关键字段：
result：可推进/待补资料/阻断blockers：阻断项needs_verify：需人工核验missing_fields：缺失字段text申请人：{applicant_id} 初筛结论：{result} 阻断项：{blockers} 待核验：{needs_verify} 缺失字段：{missing_fields}
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 20,005 | 11,304 | -43% | 1 | 1 | 0% | 3,363 | 2,320 | -31% | 0 | 0 | — |
case-01 | fail→fail | 19,719 | 18,394 | -7% | 1 | 1 | 0% | 3,202 | 3,513 | +10% | 0 | 0 | — |
case-02 | fail→pass | 18,403 | 15,628 | -15% | 1 | 1 | 0% | 2,741 | 3,198 | +17% | 0 | 0 | — |
case-03 | fail→pass | 18,152 | 15,157 | -16% | 1 | 1 | 0% | 2,658 | 3,074 | +16% | 0 | 0 | — |
case-04 | fail→pass | 14,558 | 14,558 | 0% | 1 | 1 | 0% | 1,818 | 2,844 | +56% | 0 | 0 | — |
case-05 | fail→pass | 26,840 | 20,399 | -24% | 1 | 1 | 0% | 3,890 | 3,586 | -8% | 0 | 0 | — |
case-06 | pass→pass | 11,353 | 11,529 | +2% | 1 | 1 | 0% | 1,725 | 2,341 | +36% | 0 | 0 | — |
case-08 | fail→pass | 18,512 | 15,571 | -16% | 1 | 1 | 0% | 2,394 | 2,799 | +17% | 0 | 0 | — |
case-09 | fail→pass | 12,401 | 15,145 | +22% | 1 | 1 | 0% | 1,867 | 2,498 | +34% | 0 | 0 | — |
case-10 | fail→pass | 14,310 | 11,574 | -19% | 1 | 1 | 0% | 1,802 | 2,441 | +35% | 0 | 0 | — |
case-11 | fail→pass | 8,812 | 5,964 | -32% | 1 | 1 | 0% | 1,463 | 1,507 | +3% | 0 | 0 | — |
case-12 | fail→pass | 16,233 | 10,535 | -35% | 1 | 1 | 0% | 1,990 | 2,219 | +12% | 0 | 0 | — |
case-13 | pass→pass | 25,932 | 23,230 | -10% | 1 | 1 | 0% | 3,197 | 3,668 | +15% | 0 | 0 | — |
case-14 | pass→pass | 16,042 | 11,544 | -28% | 1 | 1 | 0% | 1,923 | 2,406 | +25% | 0 | 0 | — |
case-15 | fail→pass | 12,357 | 7,658 | -38% | 1 | 1 | 0% | 1,669 | 1,734 | +4% | 0 | 0 | — |
case-16 | fail→pass | 19,235 | 13,191 | -31% | 1 | 1 | 0% | 2,690 | 2,691 | +0% | 0 | 0 | — |
case-17 | fail→pass | 18,317 | 16,002 | -13% | 1 | 1 | 0% | 2,392 | 3,042 | +27% | 0 | 0 | — |
case-18 | fail→pass | 21,084 | 5,402 | -74% | 1 | 1 | 0% | 2,957 | 1,360 | -54% | 0 | 0 | — |
case-19 | fail→pass | 16,813 | 15,986 | -5% | 1 | 1 | 0% | 2,384 | 2,818 | +18% | 0 | 0 | — |
case-20 | pass→pass | 21,856 | 29,724 | +36% | 1 | 1 | 0% | 3,124 | 3,469 | +11% | 0 | 0 | — |
case-21 | fail→pass | 20,031 | 12,766 | -36% | 1 | 1 | 0% | 2,265 | 2,696 | +19% | 0 | 0 | — |
case-22 | pass→pass | 17,032 | 15,283 | -10% | 1 | 1 | 0% | 2,590 | 2,711 | +5% | 0 | 0 | — |
case-23 | pass→pass | 22,271 | 15,189 | -32% | 1 | 1 | 0% | 2,979 | 2,912 | -2% | 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 +70 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.