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场景下对企业现金流压力做快速诊断、解释经营现金流恶化/断裂原因、形成补件与追问清单并给出下一步动作建议时触发本技能。适合尽调/审查/贷后核查前的结构化快诊与沟通提纲。
.claude/skills/aifinlab-bank-t135-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 88% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 61% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 51% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 80% | 0% |
这个 skill 用于把“现金流紧张”从一句主观感受，拆成可核验的指标口径、可解释的驱动项、可补齐的证据链，以及可执行的下一步动作。它不直接输出“能不能贷/能不能批”的终结结论，而是帮助客户经理、审查人员、贷后人员快速回答三件事：
> 配套脚本基于共享引擎输出统一结构（摘要、缺口、红旗、核验要点、补件清单、问题清单、下一步动作），便于沉淀到日报/审查要点/贷后记录。
经营现金流 vs 净利润，先回答“差在非现金项目还是营运资本占用”。最小可用输入见 input-schema.md。
建议输出结构见 output-schema.md。
scripts/run_skill.py：运行 t135 场景，生成“现金流压力诊断包”..\shared\corporate_credit_skill_engine.py：共享分析引擎（统一输出结构与规则）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | pass→pass | 9,017 | 20,843 | +131% | 1 | 1 | 0% | 1,311 | 4,380 | +234% | 0 | 0 | — |
case-01 | fail→fail | 32,278 | 50,262 | +56% | 1 | 1 | 0% | 3,982 | 6,784 | +70% | 0 | 0 | — |
case-02 | fail→fail | 28,603 | 38,463 | +34% | 1 | 1 | 0% | 3,753 | 7,018 | +87% | 0 | 0 | — |
case-03 | fail→pass | 33,401 | 40,467 | +21% | 1 | 1 | 0% | 4,194 | 7,902 | +88% | 0 | 0 | — |
case-04 | pass→pass | 11,185 | 16,419 | +47% | 1 | 1 | 0% | 1,366 | 3,768 | +176% | 0 | 0 | — |
case-05 | pass→pass | 19,109 | 25,815 | +35% | 1 | 1 | 0% | 2,611 | 5,227 | +100% | 0 | 0 | — |
case-07 | fail→fail | 28,076 | 34,862 | +24% | 1 | 1 | 0% | 3,575 | 6,613 | +85% | 0 | 0 | — |
case-08 | fail→fail | 28,665 | 34,464 | +20% | 1 | 1 | 0% | 3,681 | 6,083 | +65% | 0 | 0 | — |
case-09 | fail→pass | 22,678 | 26,920 | +19% | 1 | 1 | 0% | 3,285 | 5,034 | +53% | 0 | 0 | — |
case-10 | pass→pass | 27,929 | 39,139 | +40% | 1 | 1 | 0% | 3,577 | 6,569 | +84% | 0 | 0 | — |
case-11 | pass→pass | 32,387 | 29,876 | -8% | 1 | 1 | 0% | 4,140 | 6,056 | +46% | 0 | 0 | — |
case-12 | fail→fail | 23,225 | 37,501 | +61% | 1 | 1 | 0% | 2,886 | 6,258 | +117% | 0 | 0 | — |
case-13 | fail→pass | 23,883 | 26,509 | +11% | 1 | 1 | 0% | 3,447 | 5,540 | +61% | 0 | 0 | — |
case-14 | pass→pass | 26,827 | 42,562 | +59% | 1 | 1 | 0% | 4,061 | 7,815 | +92% | 0 | 0 | — |
case-15 | fail→pass | 30,600 | 26,300 | -14% | 1 | 1 | 0% | 3,614 | 5,471 | +51% | 0 | 0 | — |
case-16 | fail→fail | 30,397 | 31,958 | +5% | 1 | 1 | 0% | 4,644 | 5,843 | +26% | 0 | 0 | — |
case-17 | fail→fail | 26,776 | 21,770 | -19% | 1 | 1 | 0% | 3,300 | 4,858 | +47% | 0 | 0 | — |
case-18 | fail→pass | 28,482 | 36,551 | +28% | 1 | 1 | 0% | 3,634 | 6,525 | +80% | 0 | 0 | — |
case-19 | fail→pass | 26,073 | 29,187 | +12% | 1 | 1 | 0% | 3,324 | 5,632 | +69% | 0 | 0 | — |
case-20 | pass→pass | 24,464 | 30,534 | +25% | 1 | 1 | 0% | 3,501 | 6,119 | +75% | 0 | 0 | — |
case-21 | pass→pass | 31,263 | 42,840 | +37% | 1 | 1 | 0% | 4,312 | 6,992 | +62% | 0 | 0 | — |
case-22 | fail→pass | 23,719 | 26,025 | +10% | 1 | 1 | 0% | 2,936 | 5,342 | +82% | 0 | 0 | — |
case-23 | fail→fail | 30,360 | 35,782 | +18% | 1 | 1 | 0% | 4,066 | 6,618 | +63% | 0 | 0 | — |
case-24 | fail→pass | 21,849 | 36,570 | +67% | 1 | 1 | 0% | 2,971 | 6,338 | +113% | 0 | 0 | — |
case-25 | pass→pass | 25,004 | 30,730 | +23% | 1 | 1 | 0% | 3,535 | 5,980 | +69% | 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. 25 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 25 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.