Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要在银行对公金融场景下，对存量授信客户开展贷后尽调、执行偏差排查、预警信号识别、回访纪要整理或处置建议输出时使用本技能。适合服务贷后管理、风险经理和经营机构，输出结构化监测结论、补件清单、异常信号和后续动作。
.claude/skills/aifinlab-bank-t125-corporate-finance-due-diligence-post-loan-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 125% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 94% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 123% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 146% | 0% |
这个 skill 面向存量授信的贷后阶段，不做新的准入审批结论，而是帮助用户把“余额、执行、预警、回访、处置”整理成一套可跟踪的贷后动作包。它的重点不是泛泛复述贷后情况，而是尽快识别哪些问题已经发生、哪些问题仍待核验、哪些问题必须升级处置。
详细字段见 input-schema.md。
建议输出结构见 output-schema.md。
scripts/run_skill.py：读取输入 JSON，输出 Markdown 或 JSON 的贷后动作包shared/corporate_credit_skill_engine.py：本批对公授信 skill 共享的分析引擎，内含贷后场景专项规则示例调用：
bashpython scripts/run_skill.py --input input.json --format markdown
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 12,639 | 21,199 | +68% | 1 | 1 | 0% | 1,981 | 4,459 | +125% | 0 | 0 | — |
case-12 | pass→pass | 25,491 | 32,810 | +29% | 1 | 1 | 0% | 3,165 | 5,997 | +89% | 0 | 0 | — |
case-01 | fail→fail | 28,285 | 29,882 | +6% | 1 | 1 | 0% | 3,610 | 5,478 | +52% | 0 | 0 | — |
case-02 | fail→fail | 19,186 | 15,288 | -20% | 1 | 1 | 0% | 2,820 | 3,821 | +35% | 0 | 0 | — |
case-04 | fail→pass | 23,242 | 23,276 | +0% | 1 | 1 | 0% | 2,917 | 4,888 | +68% | 0 | 0 | — |
case-05 | fail→pass | 21,919 | 39,027 | +78% | 1 | 1 | 0% | 2,729 | 5,304 | +94% | 0 | 0 | — |
case-06 | pass→pass | 26,662 | 54,841 | +106% | 1 | 1 | 0% | 3,183 | 6,331 | +99% | 0 | 0 | — |
case-07 | fail→pass | 19,001 | 28,401 | +49% | 1 | 1 | 0% | 2,447 | 5,464 | +123% | 0 | 0 | — |
case-08 | pass→pass | 19,781 | 32,190 | +63% | 1 | 1 | 0% | 2,871 | 6,634 | +131% | 0 | 0 | — |
case-09 | fail→pass | 15,742 | 19,044 | +21% | 1 | 1 | 0% | 1,887 | 4,633 | +146% | 0 | 0 | — |
case-10 | fail→pass | 20,679 | 29,018 | +40% | 1 | 1 | 0% | 3,023 | 5,461 | +81% | 0 | 0 | — |
case-11 | pass→pass | 20,830 | 27,269 | +31% | 1 | 1 | 0% | 3,036 | 5,712 | +88% | 0 | 0 | — |
case-13 | pass→pass | 25,093 | 27,808 | +11% | 1 | 1 | 0% | 3,038 | 5,935 | +95% | 0 | 0 | — |
case-14 | fail→pass | 21,004 | 27,637 | +32% | 1 | 1 | 0% | 3,015 | 5,959 | +98% | 0 | 0 | — |
case-15 | pass→pass | 18,621 | 24,855 | +33% | 1 | 1 | 0% | 2,398 | 5,306 | +121% | 0 | 0 | — |
case-16 | fail→pass | 19,181 | 25,363 | +32% | 1 | 1 | 0% | 2,940 | 5,179 | +76% | 0 | 0 | — |
case-17 | pass→pass | 13,717 | 21,522 | +57% | 1 | 1 | 0% | 1,981 | 4,682 | +136% | 0 | 0 | — |
case-18 | pass→pass | 17,466 | 26,022 | +49% | 1 | 1 | 0% | 2,769 | 5,523 | +99% | 0 | 0 | — |
case-19 | pass→pass | 19,265 | 36,657 | +90% | 1 | 1 | 0% | 2,951 | 6,081 | +106% | 0 | 0 | — |
case-20 | pass→pass | 19,381 | 27,953 | +44% | 1 | 1 | 0% | 2,708 | 5,361 | +98% | 0 | 0 | — |
case-21 | pass→pass | 22,167 | 27,296 | +23% | 1 | 1 | 0% | 3,446 | 6,027 | +75% | 0 | 0 | — |
case-22 | fail→pass | 20,605 | 21,061 | +2% | 1 | 1 | 0% | 2,711 | 4,923 | +82% | 0 | 0 | — |
case-23 | pass→pass | 18,953 | 28,401 | +50% | 1 | 1 | 0% | 2,621 | 5,915 | +126% | 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 +39 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.