Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要识别保险理赔案件是否可能触发除外责任条款、分析案件中有哪些免责条款风险、结合保单和条款判断案件是否存在责任免除、梳理案件事实与免责条款的匹配关系或输出适合理赔审核与复核使用的除外责任识别意见时使用本 skill。适用于分析理赔系统案件记录、报案信息、门诊或住院病历、诊断证明、手术记录、出院小结、医疗发票与费用清单、身故证明、事故认定书、公安证明、调查记录、客服或审核备注、录音转写、OCR 文本、PDF 文档和截图转写内容,并结合产品责任、责任免除条款、等待期、既往症、自杀自伤、违法犯罪、酒驾毒驾、高风险行为、医疗必要性和其他特别约定形成结构化初步识别。
.claude/skills/aifinlab-claims-exclusion-identification-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 86% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 161% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 53% | 0% |
你是“除外责任识别助手”。
围绕“辅助识别案件是否可能触发除外责任”开展分析,重点整理案件事实、免责条款定位、条款匹配依据、争议边界、材料充分性和复核建议,帮助理赔、复核、调查、客服支持和运营人员降低除外责任漏判风险,提升案件处理一致性。
只做除外责任识别、条款匹配分析和复核支持,不替代理赔、复核、法务或调查岗位作出正式拒赔决定。
coverage-scope-judgment:适用于更通用的责任范围判断。outpatient-coverage-scope-judgment-assistant、critical-illness-coverage-scope-judgment-assistant、death-coverage-scope-judgment-assistant:适用于特定场景下的责任范围判断,不以免责条款识别为核心。以下情况优先使用本技能:
以下情况不按本技能直接处理,应提示这是更细分的问题:
写作时遵循以下规则:
输出应让理赔、复核、调查、客服支持或运营人员能快速回答:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 52,468 | 52,779 | +1% | 1 | 1 | 0% | 3,542 | 6,004 | +70% | 0 | 0 | — |
case-02 | fail→pass | 53,750 | 58,471 | +9% | 1 | 1 | 0% | 3,607 | 6,713 | +86% | 0 | 0 | — |
case-03 | fail→pass | 24,877 | 26,501 | +7% | 1 | 1 | 0% | 3,767 | 6,481 | +72% | 0 | 0 | — |
case-04 | fail→pass | 10,188 | 15,886 | +56% | 1 | 1 | 0% | 1,830 | 4,776 | +161% | 0 | 0 | — |
case-05 | fail→pass | 21,381 | 17,532 | -18% | 1 | 1 | 0% | 3,108 | 4,745 | +53% | 0 | 0 | — |
case-06 | fail→pass | 9,637 | 12,063 | +25% | 1 | 1 | 0% | 1,435 | 4,040 | +182% | 0 | 0 | — |
case-07 | pass→pass | 18,553 | 22,580 | +22% | 1 | 1 | 0% | 2,584 | 5,595 | +117% | 0 | 0 | — |
case-08 | fail→pass | 21,580 | 24,338 | +13% | 1 | 1 | 0% | 3,030 | 5,756 | +90% | 0 | 0 | — |
case-09 | fail→pass | 21,055 | 19,522 | -7% | 1 | 1 | 0% | 2,789 | 5,083 | +82% | 0 | 0 | — |
case-10 | fail→pass | 24,851 | 28,758 | +16% | 1 | 1 | 0% | 3,409 | 6,541 | +92% | 0 | 0 | — |
case-11 | fail→pass | 15,114 | 27,595 | +83% | 1 | 1 | 0% | 2,377 | 6,791 | +186% | 0 | 0 | — |
case-12 | fail→pass | 16,514 | 14,445 | -13% | 1 | 1 | 0% | 2,358 | 4,813 | +104% | 0 | 0 | — |
case-13 | fail→pass | 15,503 | 18,152 | +17% | 1 | 1 | 0% | 2,164 | 5,073 | +134% | 0 | 0 | — |
case-14 | fail→pass | 9,936 | 18,764 | +89% | 1 | 1 | 0% | 1,494 | 5,170 | +246% | 0 | 0 | — |
case-15 | fail→pass | 8,985 | 14,786 | +65% | 1 | 1 | 0% | 1,430 | 4,689 | +228% | 0 | 0 | — |
case-16 | fail→pass | 21,110 | 22,611 | +7% | 1 | 1 | 0% | 3,031 | 5,819 | +92% | 0 | 0 | — |
case-17 | fail→pass | 12,343 | 16,395 | +33% | 1 | 1 | 0% | 1,873 | 4,677 | +150% | 0 | 0 | — |
case-18 | fail→pass | 17,622 | 20,745 | +18% | 1 | 1 | 0% | 2,294 | 5,505 | +140% | 0 | 0 | — |
case-19 | pass→pass | 20,629 | 28,455 | +38% | 1 | 1 | 0% | 2,721 | 6,582 | +142% | 0 | 0 | — |
case-24 | fail→pass | 16,363 | 21,490 | +31% | 1 | 1 | 0% | 2,359 | 5,654 | +140% | 0 | 0 | — |
case-20 | fail→pass | 12,891 | 16,927 | +31% | 1 | 1 | 0% | 1,915 | 4,952 | +159% | 0 | 0 | — |
case-21 | pass→pass | 8,006 | 10,865 | +36% | 1 | 1 | 0% | 1,271 | 3,916 | +208% | 0 | 0 | — |
case-22 | fail→pass | 28,608 | 38,984 | +36% | 1 | 1 | 0% | 4,109 | 8,579 | +109% | 0 | 0 | — |
case-23 | fail→pass | 17,813 | 23,710 | +33% | 1 | 1 | 0% | 2,528 | 5,907 | +134% | 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. 24 cases were attempted. The headline lift of +88 percentage points is the difference between those two pass rates over the 24 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.