Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →经济责任审计中的廉政合规审查。审查被审计领导干部廉洁从政/从业情况， 识别利益冲突、违规违纪线索、个人事项报告等。
.claude/skills/thomasmoreai-compliance-review-jack4world/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 13% | 0% |
你是合规审计师，负责经济责任审计中的廉政合规审查。你需要以高度敏感性 识别廉政风险和违规线索，同时严格遵循"以事实为依据"的原则。
bashfor f in scope/audit-plan.md financial/findings.md decision/findings.md asset/findings.md; do ls .auditstack/$f 2>/dev/null && echo "✓ $f" || echo "✗ $f 缺失" done
汇总前序审计发现的线索，作为廉政审查的切入点。
重点关注的关联关系：
被审计人
├── 配偶、子女及其配偶
├── 其他特定关系人
└── 以上人员投资或经营的企业
│
▼
与被审计单位是否存在：
- 业务往来（采购、销售、服务）
- 资金往来（借款、担保）
- 资产交易（租赁、转让）审查方法：
| 经济行为 | 廉政风险点 | |----------|------------| | 工程招投标 | 围标串标、量身定制招标条件 | | 政府采购 | 拆分规避招标、指定品牌 | | 资产处置 | 低价转让给关联方 | | 人事任免 | 违规设岗、超编进人 | | 资金拨付 | 优亲厚友、违规审批 |
线索识别：
发现线索 → 初步核实 → 判断性质
│
┌──────────┼──────────┐
▼ ▼ ▼
管理问题 违规违纪 涉嫌违法
│ │ │
▼ ▼ ▼
审计报告 移送纪检监察 移送司法机关
中反映 机关处理 处理重要： 审计发现线索不等于定性结论。审计只负责发现和移送， 定性和处理由纪检监察或司法机关决定。措辞上用"发现...线索"、 "存在...疑点"，而非直接定性。
markdown## 廉政审查发现 [编号]：[标题] **审查事项：** [所属审查内容] **事实描述：** [客观描述，不做主观推断] **涉及人员：** [被审计人/其他相关人员] **涉及金额：** [如适用] **风险等级：** [高/中/低] **线索性质：** [管理问题/违规违纪线索/涉嫌违法线索] **建议处理：** [在审计报告中反映/移送处理/进一步核实] **审计证据：** [证据清单]
bashmkdir -p .auditstack/compliance
保存到 .auditstack/compliance/ 目录：
findings.md — 廉政审查发现clues.md — 需移送线索清单（如有）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 21,718 | 20,937 | -4% | 1 | 1 | 0% | 3,461 | 4,676 | +35% | 0 | 0 | — |
case-02 | fail→pass | 24,036 | 22,491 | -6% | 1 | 1 | 0% | 3,730 | 4,775 | +28% | 0 | 0 | — |
case-03 | fail→pass | 28,108 | 20,508 | -27% | 1 | 1 | 0% | 4,033 | 4,555 | +13% | 0 | 0 | — |
case-04 | fail→fail | 16,657 | 19,837 | +19% | 1 | 1 | 0% | 2,765 | 4,157 | +50% | 0 | 0 | — |
case-05 | fail→fail | 17,132 | 19,836 | +16% | 1 | 1 | 0% | 3,098 | 4,438 | +43% | 0 | 0 | — |
case-06 | fail→fail | 16,890 | 21,313 | +26% | 1 | 1 | 0% | 2,708 | 4,360 | +61% | 0 | 0 | — |
case-07 | fail→pass | 18,027 | 6,459 | -64% | 1 | 1 | 0% | 3,391 | 2,186 | -36% | 0 | 0 | — |
case-08 | fail→pass | 23,160 | 15,853 | -32% | 1 | 1 | 0% | 3,205 | 3,607 | +13% | 0 | 0 | — |
case-09 | fail→pass | 20,474 | 16,195 | -21% | 1 | 1 | 0% | 3,416 | 3,500 | +2% | 0 | 0 | — |
case-10 | fail→pass | 18,347 | 17,567 | -4% | 1 | 1 | 0% | 2,721 | 3,843 | +41% | 0 | 0 | — |
case-11 | pass→pass | 19,986 | 14,450 | -28% | 1 | 1 | 0% | 2,970 | 3,343 | +13% | 0 | 0 | — |
case-12 | fail→pass | 21,020 | 17,255 | -18% | 1 | 1 | 0% | 3,248 | 3,730 | +15% | 0 | 0 | — |
case-13 | fail→pass | 17,610 | 18,149 | +3% | 1 | 1 | 0% | 2,773 | 4,243 | +53% | 0 | 0 | — |
case-14 | fail→pass | 19,160 | 17,868 | -7% | 1 | 1 | 0% | 3,106 | 3,890 | +25% | 0 | 0 | — |
case-15 | fail→pass | 18,920 | 17,727 | -6% | 1 | 1 | 0% | 3,062 | 4,012 | +31% | 0 | 0 | — |
case-16 | pass→pass | 11,934 | 11,403 | -4% | 1 | 1 | 0% | 1,776 | 2,817 | +59% | 0 | 0 | — |
case-17 | fail→pass | 18,967 | 3,354 | -82% | 1 | 1 | 0% | 3,485 | 1,628 | -53% | 0 | 0 | — |
case-18 | fail→pass | 20,335 | 17,299 | -15% | 1 | 1 | 0% | 3,075 | 3,983 | +30% | 0 | 0 | — |
case-19 | pass→pass | 20,279 | 15,026 | -26% | 1 | 1 | 0% | 3,381 | 3,471 | +3% | 0 | 0 | — |
case-20 | fail→fail | 21,556 | 14,929 | -31% | 1 | 1 | 0% | 3,492 | 3,552 | +2% | 0 | 0 | — |
case-21 | fail→pass | 16,984 | 13,290 | -22% | 1 | 1 | 0% | 2,541 | 3,233 | +27% | 0 | 0 | — |
case-22 | pass→pass | 13,624 | 15,291 | +12% | 1 | 1 | 0% | 2,166 | 3,551 | +64% | 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 +64 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.