Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →面向基金合规与信息披露领域的留痕检查任务Skill，围绕「审计留痕检查助手」场景提供信息抽取、结构化分析与结果输出。
.claude/skills/aifinlab-audit-trace-check/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 76% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 42% | 0% |
bashpip install -r requirements.txt
bashpython scripts/main.py --input sample.txt --output-json result.json --output-md report.md
json{ "skill": "审计留痕检查助手", "domain": "合规与信息披露", "scene": "留痕检查", "input_summary": { "fund_code": "000000", "fund_name": "示例基金", "period": "2024Q4", "data_coverage": "净值/持仓/披露/市场" }, "key_findings": [ "关键结论1", "关键结论2" ], "data_quality": { "has_text": true, "text_length": 1200 }, "limitations": [ "仅基于输入信息形成初步判断" ] }
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 27,887 | 25,575 | -8% | 1 | 1 | 0% | 4,401 | 4,616 | +5% | 0 | 0 | — |
case-02 | fail→pass | 26,244 | 26,689 | +2% | 1 | 1 | 0% | 3,108 | 5,471 | +76% | 0 | 0 | — |
case-03 | fail→fail | 26,056 | 26,591 | +2% | 1 | 1 | 0% | 3,810 | 4,909 | +29% | 0 | 0 | — |
case-04 | fail→pass | 22,376 | 21,376 | -4% | 1 | 1 | 0% | 3,207 | 4,590 | +43% | 0 | 0 | — |
case-05 | fail→pass | 15,265 | 19,371 | +27% | 1 | 1 | 0% | 2,396 | 3,503 | +46% | 0 | 0 | — |
case-06 | fail→pass | 15,396 | 14,813 | -4% | 1 | 1 | 0% | 1,906 | 3,138 | +65% | 0 | 0 | — |
case-07 | fail→fail | 16,274 | 17,941 | +10% | 1 | 1 | 0% | 2,331 | 3,399 | +46% | 0 | 0 | — |
case-08 | fail→pass | 16,722 | 22,411 | +34% | 1 | 1 | 0% | 2,778 | 3,938 | +42% | 0 | 0 | — |
case-09 | fail→pass | 23,056 | 24,363 | +6% | 1 | 1 | 0% | 3,135 | 4,173 | +33% | 0 | 0 | — |
case-10 | fail→pass | 20,037 | 21,725 | +8% | 1 | 1 | 0% | 2,995 | 3,962 | +32% | 0 | 0 | — |
case-11 | fail→pass | 18,393 | 20,407 | +11% | 1 | 1 | 0% | 2,583 | 3,286 | +27% | 0 | 0 | — |
case-12 | fail→pass | 17,447 | 14,415 | -17% | 1 | 1 | 0% | 2,373 | 3,239 | +36% | 0 | 0 | — |
case-13 | fail→pass | 21,075 | 21,365 | +1% | 1 | 1 | 0% | 2,821 | 3,780 | +34% | 0 | 0 | — |
case-14 | fail→pass | 22,063 | 32,217 | +46% | 1 | 1 | 0% | 2,857 | 5,007 | +75% | 0 | 0 | — |
case-15 | fail→pass | 20,975 | 18,383 | -12% | 1 | 1 | 0% | 3,357 | 3,801 | +13% | 0 | 0 | — |
case-16 | fail→pass | 16,335 | 18,877 | +16% | 1 | 1 | 0% | 2,618 | 3,355 | +28% | 0 | 0 | — |
case-17 | fail→pass | 22,151 | 18,542 | -16% | 1 | 1 | 0% | 3,424 | 4,030 | +18% | 0 | 0 | — |
case-18 | fail→pass | 15,565 | 19,717 | +27% | 1 | 1 | 0% | 2,632 | 3,524 | +34% | 0 | 0 | — |
case-19 | fail→pass | 24,577 | 15,513 | -37% | 1 | 1 | 0% | 3,137 | 3,541 | +13% | 0 | 0 | — |
case-20 | pass→fail | 21,536 | 24,765 | +15% | 1 | 1 | 0% | 2,581 | 3,923 | +52% | 0 | 0 | — |
case-21 | pass→fail | 20,212 | 17,698 | -12% | 1 | 1 | 0% | 3,034 | 3,964 | +31% | 0 | 0 | — |
case-22 | pass→pass | 58,134 | 32,660 | -44% | 1 | 1 | 0% | 4,594 | 5,755 | +25% | 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. 2 cases got worse with the skill loaded, and they are 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.