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舆情异常做预警分级、识别高风险信号、形成核验要点与跟进处置动作,并输出可用于日报/周报/事件通报的结构化内容时触发本技能。适合贷后监测、授信存量管理与风险事件初筛。
.claude/skills/aifinlab-bank-t139-corporate-finance-task-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 87% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 74% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 121% | 0% |
舆情预警的目标不是“汇总新闻”,而是把分散信号转成可执行动作:先分级(严重度/可信度/扩散性),再核验(证据链与硬事件),最后给出处置(跟进动作、升级条件、沟通口径)。
这个 skill 会输出统一结构的“预警包”:摘要、信息缺口、红旗、核验要点、补件清单、沟通问题、下一步动作,便于直接落地到贷后监测、存量授信管理、风险事件通报。
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 t139 场景,生成“舆情异常预警包”..\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→pass | 28,508 | 59,123 | +107% | 1 | 1 | 0% | 4,248 | 7,946 | +87% | 0 | 0 | — |
case-02 | fail→pass | 27,776 | 28,696 | +3% | 1 | 1 | 0% | 3,474 | 5,617 | +62% | 0 | 0 | — |
case-03 | fail→pass | 29,346 | 34,020 | +16% | 1 | 1 | 0% | 3,718 | 6,478 | +74% | 0 | 0 | — |
case-04 | pass→pass | 13,260 | 23,797 | +79% | 1 | 1 | 0% | 1,891 | 4,989 | +164% | 0 | 0 | — |
case-05 | pass→pass | 20,873 | 24,258 | +16% | 1 | 1 | 0% | 2,603 | 5,134 | +97% | 0 | 0 | — |
case-19 | fail→fail | 27,881 | 34,023 | +22% | 1 | 1 | 0% | 3,553 | 6,358 | +79% | 0 | 0 | — |
case-20 | pass→pass | 21,679 | 44,473 | +105% | 1 | 1 | 0% | 2,631 | 5,007 | +90% | 0 | 0 | — |
case-06 | pass→pass | 18,473 | 14,007 | -24% | 1 | 1 | 0% | 2,941 | 3,157 | +7% | 0 | 0 | — |
case-07 | fail→pass | 25,032 | 33,953 | +36% | 1 | 1 | 0% | 3,320 | 5,657 | +70% | 0 | 0 | — |
case-08 | fail→pass | 22,607 | 37,068 | +64% | 1 | 1 | 0% | 3,161 | 6,983 | +121% | 0 | 0 | — |
case-09 | fail→pass | 17,701 | 29,619 | +67% | 1 | 1 | 0% | 2,338 | 5,670 | +143% | 0 | 0 | — |
case-10 | pass→pass | 19,112 | 28,541 | +49% | 1 | 1 | 0% | 2,407 | 4,932 | +105% | 0 | 0 | — |
case-11 | pass→pass | 18,622 | 37,733 | +103% | 1 | 1 | 0% | 2,694 | 5,701 | +112% | 0 | 0 | — |
case-12 | fail→pass | 19,269 | 36,155 | +88% | 1 | 1 | 0% | 2,914 | 5,908 | +103% | 0 | 0 | — |
case-13 | pass→pass | 21,218 | 30,316 | +43% | 1 | 1 | 0% | 3,060 | 5,243 | +71% | 0 | 0 | — |
case-14 | pass→fail | 22,062 | 37,882 | +72% | 1 | 1 | 0% | 3,084 | 6,120 | +98% | 0 | 0 | — |
case-15 | fail→pass | 24,279 | 22,459 | -7% | 1 | 1 | 0% | 3,216 | 4,753 | +48% | 0 | 0 | — |
case-16 | fail→fail | 26,238 | 33,515 | +28% | 1 | 1 | 0% | 3,338 | 5,807 | +74% | 0 | 0 | — |
case-17 | fail→fail | 21,422 | 26,489 | +24% | 1 | 1 | 0% | 2,968 | 4,920 | +66% | 0 | 0 | — |
case-18 | fail→pass | 23,151 | 34,780 | +50% | 1 | 1 | 0% | 3,149 | 5,944 | +89% | 0 | 0 | — |
case-21 | pass→pass | 25,538 | 29,925 | +17% | 1 | 1 | 0% | 3,582 | 5,835 | +63% | 0 | 0 | — |
case-22 | pass→pass | 18,679 | 26,225 | +40% | 1 | 1 | 0% | 2,614 | 5,448 | +108% | 0 | 0 | — |
case-23 | pass→pass | 24,186 | 31,340 | +30% | 1 | 1 | 0% | 2,815 | 5,312 | +89% | 0 | 0 | — |
case-24 | fail→pass | 24,876 | 40,336 | +62% | 1 | 1 | 0% | 2,920 | 6,435 | +120% | 0 | 0 | — |
case-25 | fail→pass | 23,425 | 39,242 | +68% | 1 | 1 | 0% | 3,139 | 6,366 | +103% | 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 +40 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.