Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股价差捕获/做市策略分析。当用户说"价差捕获"、"做市策略"、"spread capture"、"买卖价差交易"、"做市商"、"报价策略"时触发。基于 cn-stock-data 获取数据，分析买卖价差特征、做市收益空间、库存风险。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-spread-capture/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | 112% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 67% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 58% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 24% | 0% |
通过 cn-stock-data skill 获取数据：
# [标的] 价差捕获/做市分析报告
## 一、价差特征
| 指标 | 数值 | 日均 |
|------|------|------|
| 报价价差 | 0.05% | 0.06% |
## 二、做市收益空间
[毛利润、净利润、逆向选择成本]
## 三、库存风险
[库存偏离、半衰期、价格风险]
## 四、报价策略建议
[最优报价宽度、挂单深度]## [标的] 做市分析速览
- 报价价差 0.05%，有效价差 0.04%
- 逆向选择成本占比 35%
- 日均做市毛利空间约 2.1 万
- 库存半衰期 15 分钟，风险可控参考 references/spread-capture-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-01 | fail→fail | 46,471 | 56,713 | +22% | 1 | 1 | 0% | 6,498 | 9,358 | +44% | 0 | 0 | — |
case-02 | fail→fail | 25,863 | 14,926 | -42% | 1 | 1 | 0% | 3,692 | 3,094 | -16% | 0 | 0 | — |
case-03 | fail→fail | 35,145 | 39,764 | +13% | 1 | 1 | 0% | 5,438 | 7,216 | +33% | 0 | 0 | — |
case-04 | pass→pass | 22,403 | 30,945 | +38% | 1 | 1 | 0% | 3,140 | 5,229 | +67% | 0 | 0 | — |
case-05 | pass→pass | 26,510 | 46,404 | +75% | 1 | 1 | 0% | 5,162 | 8,178 | +58% | 0 | 0 | — |
case-06 | pass→pass | 24,661 | 19,682 | -20% | 1 | 1 | 0% | 2,918 | 3,630 | +24% | 0 | 0 | — |
case-07 | pass→pass | 8,329 | 9,337 | +12% | 1 | 1 | 0% | 1,589 | 2,336 | +47% | 0 | 0 | — |
case-08 | pass→pass | 9,180 | 7,829 | -15% | 1 | 1 | 0% | 1,531 | 2,257 | +47% | 0 | 0 | — |
case-09 | pass→pass | 7,340 | 11,919 | +62% | 1 | 1 | 0% | 1,305 | 2,808 | +115% | 0 | 0 | — |
case-10 | fail→pass | 14,738 | 27,543 | +87% | 1 | 1 | 0% | 2,011 | 4,266 | +112% | 0 | 0 | — |
case-11 | pass→pass | 8,776 | 8,638 | -2% | 1 | 1 | 0% | 1,249 | 2,253 | +80% | 0 | 0 | — |
case-12 | fail→pass | 24,977 | 24,744 | -1% | 1 | 1 | 0% | 3,704 | 4,648 | +25% | 0 | 0 | — |
case-13 | pass→pass | 23,682 | 27,008 | +14% | 1 | 1 | 0% | 3,235 | 3,989 | +23% | 0 | 0 | — |
case-14 | pass→pass | 24,212 | 31,364 | +30% | 1 | 1 | 0% | 3,160 | 4,479 | +42% | 0 | 0 | — |
case-19 | pass→pass | 24,271 | 28,404 | +17% | 1 | 1 | 0% | 3,107 | 4,700 | +51% | 0 | 0 | — |
case-15 | pass→pass | 25,895 | 25,727 | -1% | 1 | 1 | 0% | 3,524 | 4,344 | +23% | 0 | 0 | — |
case-16 | pass→pass | 27,625 | 41,429 | +50% | 1 | 1 | 0% | 4,232 | 5,934 | +40% | 0 | 0 | — |
case-17 | pass→pass | 20,540 | 43,832 | +113% | 1 | 1 | 0% | 3,097 | 6,450 | +108% | 0 | 0 | — |
case-18 | pass→pass | 23,868 | 25,235 | +6% | 1 | 1 | 0% | 3,141 | 3,808 | +21% | 0 | 0 | — |
case-20 | fail→fail | 19,162 | 54,664 | +185% | 1 | 1 | 0% | 3,066 | 4,017 | +31% | 0 | 0 | — |
case-21 | fail→fail | 33,723 | 33,752 | +0% | 1 | 1 | 0% | 4,968 | 6,561 | +32% | 0 | 0 | — |
case-22 | pass→pass | 24,699 | 19,586 | -21% | 1 | 1 | 0% | 3,201 | 3,781 | +18% | 0 | 0 | — |
case-23 | pass→pass | 24,966 | 22,187 | -11% | 1 | 1 | 0% | 3,560 | 4,496 | +26% | 0 | 0 | — |
case-24 | pass→pass | 27,182 | 25,652 | -6% | 1 | 1 | 0% | 3,599 | 4,636 | +29% | 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. 24 cases were attempted. The headline lift of +8 percentage points is the difference between those two pass rates over the 24 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.