Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when you need a bank corporate credit approval-opinion drafting assistant (审批意见/条款建议/有条件同意). Trigger this skill for把授信申请要素、风险点与缓释措施整理成可会签的审批意见草稿、提款前条件与贷后监测条款，并可用脚本从结构化输入生成 Markdown 审批意见初稿。
.claude/skills/aifinlab-bank-t144-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 90% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 80% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 47% | 0% |
本技能用于快速产出“审批会/会签可用”的审批意见草稿，重点不是替代审批，而是：
输出适合：审查意见初稿、审批会准备材料、条款清单整理、会签沟通底稿。
最低必需（详见 references/input-schema.md）：
company.nameapplication.product_type / application.amount_mn / application.term / application.purposerisk_points[]（若缺失，则只能输出框架）建议提供：
mitigants[]：风险缓释措施（条款/流程/监测）proposed_terms[]：条款建议或已初步拟定条款gaps[]：明确缺口，脚本会纳入意见草稿脚本输出字段见 references/output-schema.md。
bashpython scripts/run_skill.py --input assets/example-input.json --format markdown python scripts/run_skill.py --input assets/example-input.json --format json
scripts/run_skill.py：命令行入口scripts/approval_opinion_drafter.py：t144 场景封装（调用共享引擎）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 26,900 | 39,751 | +48% | 1 | 1 | 0% | 4,081 | 5,189 | +27% | 0 | 0 | — |
case-02 | fail→pass | 18,034 | 25,065 | +39% | 1 | 1 | 0% | 2,739 | 5,206 | +90% | 0 | 0 | — |
case-03 | pass→pass | 9,163 | 18,508 | +102% | 1 | 1 | 0% | 1,099 | 3,392 | +209% | 0 | 0 | — |
case-04 | fail→pass | 26,230 | 25,006 | -5% | 1 | 1 | 0% | 4,225 | 4,224 | -0% | 0 | 0 | — |
case-05 | pass→pass | 20,217 | 24,839 | +23% | 1 | 1 | 0% | 2,707 | 4,475 | +65% | 0 | 0 | — |
case-06 | pass→pass | 15,499 | 30,856 | +99% | 1 | 1 | 0% | 2,226 | 5,203 | +134% | 0 | 0 | — |
case-07 | pass→pass | 14,171 | 20,032 | +41% | 1 | 1 | 0% | 2,167 | 3,795 | +75% | 0 | 0 | — |
case-08 | pass→fail | 17,799 | 23,622 | +33% | 1 | 1 | 0% | 2,650 | 4,226 | +59% | 0 | 0 | — |
case-09 | pass→pass | 23,861 | 34,705 | +45% | 1 | 1 | 0% | 3,099 | 6,364 | +105% | 0 | 0 | — |
case-10 | fail→pass | 17,954 | 27,022 | +51% | 1 | 1 | 0% | 2,579 | 5,359 | +108% | 0 | 0 | — |
case-11 | pass→pass | 21,868 | 25,430 | +16% | 1 | 1 | 0% | 2,874 | 4,864 | +69% | 0 | 0 | — |
case-12 | pass→pass | 15,788 | 27,204 | +72% | 1 | 1 | 0% | 2,426 | 4,638 | +91% | 0 | 0 | — |
case-13 | fail→pass | 22,412 | 25,711 | +15% | 1 | 1 | 0% | 2,875 | 5,171 | +80% | 0 | 0 | — |
case-14 | fail→pass | 22,424 | 29,455 | +31% | 1 | 1 | 0% | 3,430 | 5,054 | +47% | 0 | 0 | — |
case-15 | pass→pass | 19,090 | 23,653 | +24% | 1 | 1 | 0% | 2,333 | 4,959 | +113% | 0 | 0 | — |
case-16 | pass→pass | 20,032 | 27,222 | +36% | 1 | 1 | 0% | 2,857 | 4,600 | +61% | 0 | 0 | — |
case-17 | pass→pass | 20,635 | 25,163 | +22% | 1 | 1 | 0% | 2,565 | 4,643 | +81% | 0 | 0 | — |
case-18 | pass→pass | 21,589 | 27,627 | +28% | 1 | 1 | 0% | 2,890 | 4,689 | +62% | 0 | 0 | — |
case-19 | pass→pass | 19,004 | 29,430 | +55% | 1 | 1 | 0% | 2,842 | 5,779 | +103% | 0 | 0 | — |
case-20 | pass→pass | 22,560 | 25,520 | +13% | 1 | 1 | 0% | 3,099 | 4,483 | +45% | 0 | 0 | — |
case-21 | pass→pass | 16,827 | 20,740 | +23% | 1 | 1 | 0% | 2,902 | 4,329 | +49% | 0 | 0 | — |
case-22 | pass→pass | 17,931 | 42,300 | +136% | 1 | 1 | 0% | 2,480 | 4,518 | +82% | 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. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 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.