Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股暗池/大宗交易策略。当用户说"暗池"、"dark pool"、"大宗交易"、"大宗减持"、"盘后大宗"、"协议转让"、"大宗折价"时触发。基于 cn-stock-data 获取数据，分析大宗交易机会与策略。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-dark-pool/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-18 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-23 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -47% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 14% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 35% | 0% |
通过 cn-stock-data skill 获取数据：
# [标的] 大宗交易分析报告
## 一、大宗交易概览
| 日期 | 价格 | 折价 | 量(万股) | 买方 |
|------|------|------|---------|------|
## 二、信号解读
[折价原因、买卖方分析]
## 三、策略建议
[接盘/回避建议]
## 四、风险提示## [标的] 大宗速览
- 近5日大宗成交3笔，累计500万股
- 平均折价 -6.5%
- 买方：机构专用席位
- 信号：机构折价接盘，中性偏正面参考 references/dark-pool-guide.md 获取详细方法论与 A股实证研究。
python# 调用 skill result = run_skill({ "param1": "value1", "param2": "value2" })
bashpython scripts/run_skill.py --input data.json
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-18 | fail→pass | 16,708 | 4,905 | -71% | 1 | 1 | 0% | 2,141 | 1,271 | -41% | 0 | 0 | — |
case-01 | fail→fail | 32,392 | 23,359 | -28% | 1 | 1 | 0% | 4,237 | 4,265 | +1% | 0 | 0 | — |
case-02 | fail→fail | 15,312 | 11,172 | -27% | 1 | 1 | 0% | 2,237 | 2,073 | -7% | 0 | 0 | — |
case-03 | fail→fail | 28,946 | 31,031 | +7% | 1 | 1 | 0% | 3,677 | 4,608 | +25% | 0 | 0 | — |
case-04 | pass→pass | 18,234 | 13,936 | -24% | 1 | 1 | 0% | 2,388 | 2,729 | +14% | 0 | 0 | — |
case-05 | pass→pass | 21,792 | 23,248 | +7% | 1 | 1 | 0% | 3,084 | 4,171 | +35% | 0 | 0 | — |
case-06 | fail→fail | 24,071 | 28,578 | +19% | 1 | 1 | 0% | 2,900 | 4,528 | +56% | 0 | 0 | — |
case-07 | pass→pass | 16,255 | 22,142 | +36% | 1 | 1 | 0% | 2,201 | 3,304 | +50% | 0 | 0 | — |
case-08 | fail→fail | 16,403 | 10,355 | -37% | 1 | 1 | 0% | 2,228 | 1,391 | -38% | 0 | 0 | — |
case-09 | fail→fail | 24,516 | 25,747 | +5% | 1 | 1 | 0% | 3,393 | 3,868 | +14% | 0 | 0 | — |
case-10 | pass→pass | 22,831 | 21,912 | -4% | 1 | 1 | 0% | 3,276 | 3,769 | +15% | 0 | 0 | — |
case-11 | fail→fail | 21,243 | 23,923 | +13% | 1 | 1 | 0% | 2,986 | 4,224 | +41% | 0 | 0 | — |
case-12 | pass→pass | 26,167 | 45,876 | +75% | 1 | 1 | 0% | 3,751 | 4,968 | +32% | 0 | 0 | — |
case-13 | pass→pass | 18,593 | 11,304 | -39% | 1 | 1 | 0% | 2,572 | 2,292 | -11% | 0 | 0 | — |
case-14 | pass→pass | 24,910 | 5,698 | -77% | 1 | 1 | 0% | 2,323 | 1,498 | -36% | 0 | 0 | — |
case-15 | pass→fail | 37,513 | 20,315 | -46% | 1 | 1 | 0% | 3,914 | 2,090 | -47% | 0 | 0 | — |
case-16 | pass→pass | 25,286 | 38,387 | +52% | 1 | 1 | 0% | 3,503 | 4,250 | +21% | 0 | 0 | — |
case-17 | pass→pass | 21,883 | 30,977 | +42% | 1 | 1 | 0% | 3,776 | 4,967 | +32% | 0 | 0 | — |
case-19 | pass→pass | 25,509 | 29,164 | +14% | 1 | 1 | 0% | 3,307 | 3,981 | +20% | 0 | 0 | — |
case-20 | pass→pass | 10,187 | 7,896 | -22% | 1 | 1 | 0% | 1,585 | 1,943 | +23% | 0 | 0 | — |
case-21 | pass→pass | 15,403 | 14,419 | -6% | 1 | 1 | 0% | 2,177 | 2,847 | +31% | 0 | 0 | — |
case-22 | pass→pass | 20,885 | 20,075 | -4% | 1 | 1 | 0% | 3,064 | 3,715 | +21% | 0 | 0 | — |
case-23 | fail→pass | 22,660 | 27,112 | +20% | 1 | 1 | 0% | 2,798 | 2,771 | -1% | 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, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +4 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is 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.