Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要梳理保险理赔案件的事实时间线、重建出险到结案的事件顺序、整合病历报案材料审核记录中的时间信息、识别时间冲突或生成适合理赔审核调查复核客服使用的标准化时间轴时使用本 skill。适用于汇总理赔系统案件记录、报案记录、出险说明、门急诊记录、住院记录、手术记录、检查检验报告、病历首页、出院小结、发票或费用清单、补件记录、审核记录、调查记录、结案记录、客服工单记录、沟通备注、OCR 文本、PDF 文档和截图转写内容,提炼关键时间节点、阶段划分、时间间隔、时间冲突和后续核查建议。
.claude/skills/aifinlab-claims-fact-timeline-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 113% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 69% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 47% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 111% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 62% | 0% |
你是“案件事实时间线助手”。
围绕“重建案件事实时间线”开展整理,重点提炼案件基本信息、关键事件顺序、多来源时间信息对齐、关键时间间隔、时间冲突、阶段划分和后续核查建议,帮助理赔、客服、调查、复核和争议处理人员快速理解案件来龙去脉。
只做案件事实时间线整理、节点提取、冲突提示和流程辅助判断,不替代理赔、调查、复核或法务岗位作出正式责任结论。
claims-case-summary:适用于通用案件摘要。claims-case-summary-audit-assistant:适用于审核争议和证据摘要。claims-case-summary-customer-service-assistant:适用于客服进度说明和沟通口径。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:
除非用户另有要求,严格按 references/output-schema.md 的顺序输出。
写作时遵循以下规则:
输出应让理赔、调查、客服或复核人员能快速回答:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 25,378 | 37,028 | +46% | 1 | 1 | 0% | 4,274 | 7,905 | +85% | 0 | 0 | — |
case-02 | pass→pass | 36,594 | 15,149 | -59% | 1 | 1 | 0% | 5,902 | 4,790 | -19% | 0 | 0 | — |
case-18 | pass→pass | 13,246 | 20,736 | +57% | 1 | 1 | 0% | 1,983 | 5,359 | +170% | 0 | 0 | — |
case-19 | pass→pass | 14,226 | 13,019 | -8% | 1 | 1 | 0% | 2,306 | 4,382 | +90% | 0 | 0 | — |
case-20 | pass→pass | 19,329 | 20,352 | +5% | 1 | 1 | 0% | 2,956 | 5,574 | +89% | 0 | 0 | — |
case-03 | fail→pass | 13,505 | 15,246 | +13% | 1 | 1 | 0% | 2,097 | 4,472 | +113% | 0 | 0 | — |
case-04 | fail→pass | 17,220 | 12,479 | -28% | 1 | 1 | 0% | 2,359 | 3,984 | +69% | 0 | 0 | — |
case-05 | fail→pass | 14,899 | 6,528 | -56% | 1 | 1 | 0% | 2,120 | 3,107 | +47% | 0 | 0 | — |
case-06 | pass→pass | 14,031 | 15,127 | +8% | 1 | 1 | 0% | 2,015 | 4,729 | +135% | 0 | 0 | — |
case-07 | pass→pass | 13,696 | 11,364 | -17% | 1 | 1 | 0% | 2,204 | 4,011 | +82% | 0 | 0 | — |
case-08 | pass→pass | 17,333 | 18,296 | +6% | 1 | 1 | 0% | 2,516 | 5,329 | +112% | 0 | 0 | — |
case-09 | fail→fail | 10,367 | 15,441 | +49% | 1 | 1 | 0% | 1,849 | 5,019 | +171% | 0 | 0 | — |
case-10 | fail→pass | 14,568 | 16,287 | +12% | 1 | 1 | 0% | 2,501 | 5,288 | +111% | 0 | 0 | — |
case-11 | pass→pass | 14,485 | 16,202 | +12% | 1 | 1 | 0% | 2,099 | 4,908 | +134% | 0 | 0 | — |
case-12 | fail→pass | 16,797 | 12,860 | -23% | 1 | 1 | 0% | 2,664 | 4,315 | +62% | 0 | 0 | — |
case-13 | pass→pass | 18,141 | 22,469 | +24% | 1 | 1 | 0% | 2,984 | 6,264 | +110% | 0 | 0 | — |
case-14 | pass→pass | 17,488 | 17,070 | -2% | 1 | 1 | 0% | 2,747 | 5,023 | +83% | 0 | 0 | — |
case-15 | pass→pass | 12,145 | 19,379 | +60% | 1 | 1 | 0% | 1,982 | 5,648 | +185% | 0 | 0 | — |
case-16 | pass→pass | 13,098 | 8,498 | -35% | 1 | 1 | 0% | 2,117 | 3,603 | +70% | 0 | 0 | — |
case-17 | fail→pass | 21,908 | 26,663 | +22% | 1 | 1 | 0% | 3,088 | 6,276 | +103% | 0 | 0 | — |
case-21 | pass→pass | 16,704 | 12,435 | -26% | 1 | 1 | 0% | 2,601 | 4,185 | +61% | 0 | 0 | — |
case-22 | fail→pass | 14,732 | 10,611 | -28% | 1 | 1 | 0% | 2,353 | 3,823 | +62% | 0 | 0 | — |
case-23 | pass→pass | 22,037 | 17,240 | -22% | 1 | 1 | 0% | 3,129 | 4,636 | +48% | 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 +30 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.