Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要在银行交易银行与普惠场景下，围绕产品推荐进行适配、配置、推荐、方案设计或备选方案比较时使用本技能。适合输出主方案、备选方案、匹配逻辑、风险提示和沟通要点。
.claude/skills/aifinlab-bank-t243-transaction-banking-inclusive-finance-product-recommendation-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 52% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 208% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 195% | 0% |
本技能面向交易银行与普惠金融场景，既要兼顾准入效率和材料真实性，也要把结算、现金管理、供应链和贸易融资方案写到能落地的层面。 当前这支 skill 更偏向服务 交易银行和普惠团队，输出时要特别注意 兼顾准入效率、方案落地和风险边界。
本技能提供产品匹配与推荐排序脚本，适合批量客户的快速筛选与方案生成。
scripts/product_recommendation_builder.py用途：基于客户需求与约束，对产品池进行匹配评分并生成推荐方案。
输入字段（JSON）：
client_profile: 客户画像needs: 需求与约束product_catalog: 产品池（适用范围、门槛、费率）rules: 匹配权重与禁用条件输出内容：
命令行示例：
python scripts/product_recommendation_builder.py --input assets/product_input.json --output outputs/product_recommendation.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-12 | fail→pass | 8,512 | 2,827 | -67% | 1 | 1 | 0% | 1,257 | 1,507 | +20% | 0 | 0 | — |
case-01 | fail→pass | 63,232 | 64,451 | +2% | 1 | 1 | 0% | 4,979 | 6,505 | +31% | 0 | 0 | — |
case-02 | fail→fail | 26,865 | 32,495 | +21% | 1 | 1 | 0% | 3,796 | 5,743 | +51% | 0 | 0 | — |
case-03 | fail→pass | 60,170 | 67,645 | +12% | 1 | 1 | 0% | 4,345 | 6,613 | +52% | 0 | 0 | — |
case-04 | pass→pass | 15,678 | 22,681 | +45% | 1 | 1 | 0% | 2,261 | 4,025 | +78% | 0 | 0 | — |
case-05 | pass→pass | 6,564 | 9,555 | +46% | 1 | 1 | 0% | 910 | 2,332 | +156% | 0 | 0 | — |
case-06 | fail→pass | 8,726 | 19,682 | +126% | 1 | 1 | 0% | 1,265 | 3,890 | +208% | 0 | 0 | — |
case-07 | pass→pass | 21,472 | 23,197 | +8% | 1 | 1 | 0% | 2,829 | 4,507 | +59% | 0 | 0 | — |
case-08 | fail→fail | 25,110 | 26,518 | +6% | 1 | 1 | 0% | 3,733 | 5,092 | +36% | 0 | 0 | — |
case-09 | pass→pass | 18,379 | 25,243 | +37% | 1 | 1 | 0% | 2,945 | 5,035 | +71% | 0 | 0 | — |
case-10 | pass→pass | 26,685 | 34,434 | +29% | 1 | 1 | 0% | 4,039 | 6,176 | +53% | 0 | 0 | — |
case-11 | pass→pass | 23,505 | 27,806 | +18% | 1 | 1 | 0% | 3,483 | 4,971 | +43% | 0 | 0 | — |
case-13 | fail→pass | 27,617 | 13,784 | -50% | 1 | 1 | 0% | 1,263 | 3,731 | +195% | 0 | 0 | — |
case-14 | pass→pass | 17,459 | 22,657 | +30% | 1 | 1 | 0% | 2,560 | 4,082 | +59% | 0 | 0 | — |
case-15 | pass→pass | 20,530 | 30,064 | +46% | 1 | 1 | 0% | 2,706 | 5,138 | +90% | 0 | 0 | — |
case-16 | pass→pass | 23,825 | 27,405 | +15% | 1 | 1 | 0% | 3,246 | 4,703 | +45% | 0 | 0 | — |
case-17 | pass→pass | 23,790 | 28,089 | +18% | 1 | 1 | 0% | 3,380 | 5,046 | +49% | 0 | 0 | — |
case-18 | pass→pass | 24,973 | 35,045 | +40% | 1 | 1 | 0% | 3,386 | 5,883 | +74% | 0 | 0 | — |
case-19 | fail→pass | 23,368 | 27,203 | +16% | 1 | 1 | 0% | 3,394 | 5,113 | +51% | 0 | 0 | — |
case-20 | pass→pass | 21,892 | 33,384 | +52% | 1 | 1 | 0% | 3,279 | 5,872 | +79% | 0 | 0 | — |
case-21 | pass→fail | 20,825 | 26,600 | +28% | 1 | 1 | 0% | 3,045 | 4,998 | +64% | 0 | 0 | — |
case-22 | pass→pass | 24,267 | 26,069 | +7% | 1 | 1 | 0% | 3,536 | 4,688 | +33% | 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, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +23 percentage points is the difference between those two pass rates over the 21 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.