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关联方与关联交易做穿透分析（界定边界、归类交易、做合理性测试、穿透资金往来与利润转移、识别风险并形成补件与核验清单）时触发本技能。适合授信尽调/审查/贷后排查的结构化穿透输出。
.claude/skills/aifinlab-bank-t138-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 77% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 81% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 120% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 148% | 0% |
关联交易穿透不是“把关联方名单抄一遍”，而是回答三件事：
本 skill 输出的是“可核验、可执行”的穿透包：摘要、缺口、红旗、核验要点、补件清单、访谈问题与下一步动作。它不替代审计/法务结论，也不把穿透结论写成确定性违规定性。
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 t138 场景，生成“关联交易穿透包”..\shared\corporate_credit_skill_engine.py：共享引擎（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-01 | fail→fail | 32,344 | 53,666 | +66% | 1 | 1 | 0% | 4,250 | 7,026 | +65% | 0 | 0 | — |
case-02 | fail→fail | 29,377 | 40,567 | +38% | 1 | 1 | 0% | 4,058 | 6,776 | +67% | 0 | 0 | — |
case-03 | pass→pass | 7,964 | 8,762 | +10% | 1 | 1 | 0% | 1,106 | 2,747 | +148% | 0 | 0 | — |
case-04 | pass→pass | 21,183 | 26,512 | +25% | 1 | 1 | 0% | 2,646 | 4,857 | +84% | 0 | 0 | — |
case-05 | pass→pass | 16,158 | 18,423 | +14% | 1 | 1 | 0% | 2,252 | 4,278 | +90% | 0 | 0 | — |
case-06 | fail→pass | 21,246 | 19,807 | -7% | 1 | 1 | 0% | 2,620 | 4,325 | +65% | 0 | 0 | — |
case-07 | pass→pass | 19,604 | 34,939 | +78% | 1 | 1 | 0% | 2,928 | 6,168 | +111% | 0 | 0 | — |
case-21 | fail→pass | 30,981 | 25,757 | -17% | 1 | 1 | 0% | 2,757 | 4,881 | +77% | 0 | 0 | — |
case-08 | pass→pass | 23,681 | 32,708 | +38% | 1 | 1 | 0% | 3,416 | 5,834 | +71% | 0 | 0 | — |
case-09 | pass→pass | 22,774 | 28,459 | +25% | 1 | 1 | 0% | 2,837 | 5,781 | +104% | 0 | 0 | — |
case-10 | pass→pass | 24,867 | 37,746 | +52% | 1 | 1 | 0% | 3,646 | 6,394 | +75% | 0 | 0 | — |
case-11 | pass→pass | 22,609 | 29,067 | +29% | 1 | 1 | 0% | 3,257 | 5,839 | +79% | 0 | 0 | — |
case-12 | pass→pass | 21,828 | 31,109 | +43% | 1 | 1 | 0% | 3,310 | 5,667 | +71% | 0 | 0 | — |
case-13 | fail→pass | 22,101 | 32,963 | +49% | 1 | 1 | 0% | 3,189 | 5,771 | +81% | 0 | 0 | — |
case-14 | pass→pass | 19,465 | 35,149 | +81% | 1 | 1 | 0% | 2,970 | 6,081 | +105% | 0 | 0 | — |
case-15 | pass→pass | 26,830 | 24,526 | -9% | 1 | 1 | 0% | 3,417 | 5,130 | +50% | 0 | 0 | — |
case-16 | pass→pass | 21,047 | 28,898 | +37% | 1 | 1 | 0% | 2,445 | 5,559 | +127% | 0 | 0 | — |
case-17 | pass→pass | 23,381 | 30,148 | +29% | 1 | 1 | 0% | 2,990 | 5,341 | +79% | 0 | 0 | — |
case-18 | pass→pass | 31,604 | 48,092 | +52% | 1 | 1 | 0% | 4,178 | 6,877 | +65% | 0 | 0 | — |
case-19 | pass→pass | 21,669 | 36,014 | +66% | 1 | 1 | 0% | 2,620 | 5,864 | +124% | 0 | 0 | — |
case-20 | fail→pass | 23,341 | 43,357 | +86% | 1 | 1 | 0% | 3,496 | 7,693 | +120% | 0 | 0 | — |
case-22 | pass→pass | 20,364 | 29,493 | +45% | 1 | 1 | 0% | 3,127 | 5,881 | +88% | 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.
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.