Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →从调查日志起草或更新调查备忘录。当调查进展到可以撰写第一版备忘录时， 或当新数据已添加且现有草案需要更新时使用。
.claude/skills/thomasmoreai-investigation-memo-zhou210712/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 78% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 22% | 0% |
从调查日志起草第一版调查备忘录，或当新数据已添加时更新现有草案。
markdown保密 · 内部调查工作成果 # 调查备忘录：[调查事项名称] **调查期间：** [开始日期] — [日期] **调查人员：** [姓名/角色] --- ## 一、投诉/举报概述 [投诉来源、时间、内容摘要] ## 二、调查过程 [取证情况、访谈人员及日期] ## 三、事实发现 [按时间线或争议焦点组织，每个发现附证据引用] ## 四、证据分析 [证据的三性审查——真实性、合法性、关联性] [证据之间的矛盾和印证关系] ## 五、初步结论 [基于现有证据的事实认定] [存在的证据缺口和不确定性] ## 六、建议 [处理建议及法律依据]
保密说明（《律师法》第38条）。 本备忘录包含律师在执业活动中知悉的保密信息，受《律师法》第38条保密义务保护。[法条原文] 在中国法下不存在独立的"attorney work product"保护制度，法院有权依法调取相关材料。建议将本备忘录与外部律师的法律意见书分开管理。
/employment-legal:investigation-memo [调查事项名称]/employment-legal:investigation-memo [调查事项名称]
（如已有草案则更新）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 22,472 | 20,789 | -7% | 1 | 1 | 0% | 3,323 | 3,634 | +9% | 0 | 0 | — |
case-02 | fail→pass | 27,083 | 19,819 | -27% | 1 | 1 | 0% | 2,931 | 3,644 | +24% | 0 | 0 | — |
case-03 | fail→pass | 26,670 | 25,346 | -5% | 1 | 1 | 0% | 4,050 | 4,429 | +9% | 0 | 0 | — |
case-04 | pass→pass | 16,484 | 16,527 | +0% | 1 | 1 | 0% | 2,760 | 3,040 | +10% | 0 | 0 | — |
case-05 | pass→pass | 21,861 | 19,920 | -9% | 1 | 1 | 0% | 3,415 | 3,613 | +6% | 0 | 0 | — |
case-06 | pass→pass | 26,236 | 28,132 | +7% | 1 | 1 | 0% | 3,714 | 4,685 | +26% | 0 | 0 | — |
case-07 | pass→pass | 19,712 | 19,391 | -2% | 1 | 1 | 0% | 2,952 | 3,519 | +19% | 0 | 0 | — |
case-08 | fail→pass | 14,571 | 21,309 | +46% | 1 | 1 | 0% | 2,125 | 3,783 | +78% | 0 | 0 | — |
case-09 | fail→pass | 26,936 | 22,965 | -15% | 1 | 1 | 0% | 3,308 | 4,051 | +22% | 0 | 0 | — |
case-10 | pass→pass | 20,696 | 14,791 | -29% | 1 | 1 | 0% | 3,119 | 2,721 | -13% | 0 | 0 | — |
case-11 | pass→pass | 21,725 | 21,810 | +0% | 1 | 1 | 0% | 3,255 | 3,915 | +20% | 0 | 0 | — |
case-12 | fail→pass | 21,243 | 22,025 | +4% | 1 | 1 | 0% | 3,292 | 3,933 | +19% | 0 | 0 | — |
case-13 | fail→fail | 23,362 | 19,794 | -15% | 1 | 1 | 0% | 3,122 | 3,544 | +14% | 0 | 0 | — |
case-14 | fail→pass | 14,391 | 20,233 | +41% | 1 | 1 | 0% | 2,130 | 3,571 | +68% | 0 | 0 | — |
case-15 | fail→pass | 20,866 | 24,119 | +16% | 1 | 1 | 0% | 3,059 | 4,108 | +34% | 0 | 0 | — |
case-16 | fail→pass | 14,006 | 18,022 | +29% | 1 | 1 | 0% | 2,217 | 3,380 | +52% | 0 | 0 | — |
case-17 | fail→pass | 22,875 | 19,394 | -15% | 1 | 1 | 0% | 3,405 | 3,228 | -5% | 0 | 0 | — |
case-18 | fail→pass | 18,689 | 18,469 | -1% | 1 | 1 | 0% | 3,052 | 3,333 | +9% | 0 | 0 | — |
case-19 | fail→pass | 22,134 | 23,336 | +5% | 1 | 1 | 0% | 3,252 | 3,973 | +22% | 0 | 0 | — |
case-20 | fail→pass | 21,173 | 19,149 | -10% | 1 | 1 | 0% | 3,188 | 3,299 | +3% | 0 | 0 | — |
case-21 | fail→pass | 15,447 | 20,170 | +31% | 1 | 1 | 0% | 2,154 | 3,920 | +82% | 0 | 0 | — |
case-22 | fail→pass | 17,222 | 23,372 | +36% | 1 | 1 | 0% | 2,655 | 3,527 | +33% | 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 +68 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.