Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when you need a bank corporate-risk monitoring assistant for enterprise litigation/penalty/enforcement scanning (处罚/诉讼/被执行/失信等). Trigger this skill for requests to做“扫描分级、核验清单、处置动作、升级路径”的结构化输出，并可用脚本把输入事件台账生成监测简报。
.claude/skills/aifinlab-bank-t140-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 87% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 54% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 74% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 102% | 0% |
本技能用于“把外部负面信号变成可推进动作”的监测工具：对企业诉讼、仲裁、行政处罚、监管措施、被执行、失信/限高等信息做分级与核验，输出监测简报（摘要 + 事件节选 + 缺口 + 核验要点 + 动作建议 + 升级条件）。
定位强调三件事：
典型输入：
典型输出：
最低必需（建议结构化，详见 references/input-schema.md）：
company.namecompany.uscc（或 company.registration_no）monitoring.time_windowevents[]（至少包含 date、event_type、role、status、amount_mn/amount_text、counterparty）可选增强：
monitoring.sources：来源列表与口径说明monitoring.baseline：历史基线（例如过去12个月新增案件数）exposures：授信敞口、担保、押品信息（用于判断“同量级”）脚本输出（JSON/Markdown）字段见 references/output-schema.md。
当你已经把事件整理成结构化 JSON（哪怕不完整），可以直接用脚本生成监测简报：
bashpython scripts/run_skill.py --input assets/example-input.json --format markdown python scripts/run_skill.py --input assets/example-input.json --format json
脚本说明：
scripts/run_skill.py：命令行入口（读取 JSON -> 生成 Markdown/JSON）scripts/penalty_litigation_scan.py：t140 场景封装（调用共享引擎 shared/corporate_finance_task_engine.py）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-14 | fail→pass | 9,301 | 4,761 | -49% | 1 | 1 | 0% | 1,649 | 2,330 | +41% | 0 | 0 | — |
case-01 | fail→fail | 27,704 | 46,761 | +69% | 1 | 1 | 0% | 3,799 | 6,315 | +66% | 0 | 0 | — |
case-02 | fail→pass | 17,449 | 20,671 | +18% | 1 | 1 | 0% | 2,522 | 4,723 | +87% | 0 | 0 | — |
case-03 | fail→pass | 12,276 | 12,409 | +1% | 1 | 1 | 0% | 2,113 | 3,253 | +54% | 0 | 0 | — |
case-04 | fail→pass | 19,151 | 22,817 | +19% | 1 | 1 | 0% | 2,534 | 4,409 | +74% | 0 | 0 | — |
case-05 | fail→pass | 18,973 | 34,349 | +81% | 1 | 1 | 0% | 2,931 | 5,911 | +102% | 0 | 0 | — |
case-06 | pass→pass | 25,625 | 30,418 | +19% | 1 | 1 | 0% | 3,263 | 6,513 | +100% | 0 | 0 | — |
case-07 | pass→pass | 15,639 | 22,739 | +45% | 1 | 1 | 0% | 2,514 | 5,321 | +112% | 0 | 0 | — |
case-08 | fail→pass | 22,324 | 29,689 | +33% | 1 | 1 | 0% | 3,132 | 5,268 | +68% | 0 | 0 | — |
case-09 | fail→fail | 17,572 | 28,723 | +63% | 1 | 1 | 0% | 2,527 | 5,967 | +136% | 0 | 0 | — |
case-10 | fail→pass | 25,211 | 24,790 | -2% | 1 | 1 | 0% | 3,641 | 4,824 | +32% | 0 | 0 | — |
case-11 | fail→pass | 61,585 | 36,358 | -41% | 1 | 1 | 0% | 3,344 | 6,337 | +90% | 0 | 0 | — |
case-12 | pass→pass | 39,620 | 34,025 | -14% | 1 | 1 | 0% | 2,751 | 5,939 | +116% | 0 | 0 | — |
case-13 | fail→pass | 21,557 | 25,050 | +16% | 1 | 1 | 0% | 3,388 | 5,009 | +48% | 0 | 0 | — |
case-15 | fail→fail | 22,331 | 22,195 | -1% | 1 | 1 | 0% | 3,103 | 5,086 | +64% | 0 | 0 | — |
case-16 | pass→pass | 20,753 | 27,584 | +33% | 1 | 1 | 0% | 2,125 | 5,038 | +137% | 0 | 0 | — |
case-17 | fail→pass | 16,693 | 25,629 | +54% | 1 | 1 | 0% | 2,619 | 5,316 | +103% | 0 | 0 | — |
case-18 | pass→pass | 15,268 | 23,952 | +57% | 1 | 1 | 0% | 2,411 | 4,542 | +88% | 0 | 0 | — |
case-19 | fail→fail | 28,771 | 34,030 | +18% | 1 | 1 | 0% | 3,509 | 6,075 | +73% | 0 | 0 | — |
case-20 | fail→pass | 9,002 | 4,423 | -51% | 1 | 1 | 0% | 1,600 | 2,098 | +31% | 0 | 0 | — |
case-21 | pass→pass | 24,889 | 24,722 | -1% | 1 | 1 | 0% | 2,959 | 4,786 | +62% | 0 | 0 | — |
case-22 | fail→fail | 18,507 | 22,492 | +22% | 1 | 1 | 0% | 2,534 | 5,110 | +102% | 0 | 0 | — |
case-23 | fail→fail | 19,218 | 28,180 | +47% | 1 | 1 | 0% | 2,830 | 5,832 | +106% | 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 +48 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.