Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股高频微观结构/盘口深度分析。当用户说"高频微观结构"、"盘口深度"、"Level2深度"、"逐笔委托"、"高频盘口"、"微结构分析"时触发。基于 cn-stock-data 获取数据，分析盘口深度、有效价差、知情交易概率。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-hft-microstructure/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 49% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 46% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 77% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 17% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 26% | 0% |
通过 cn-stock-data skill 获取数据：
# [标的] 高频微观结构分析报告
## 一、盘口深度概览
| 指标 | 数值 | 分位数 | 评估 |
|------|------|--------|------|
| OIR | +0.15 | P62 | 买盘略占优 |
| 有效价差 | 0.08% | P35 | 流动性良好 |
## 二、知情交易分析
[PIN/VPIN 估算结果，信息不对称评估]
## 三、价格冲击
[Kyle Lambda，冲击成本估算]
## 四、异常信号
[盘口异常事件汇总]
## 五、交易建议
[基于微结构的最优执行建议]## [标的] 微结构速览
- OIR +0.15，买盘略占优
- 有效价差 0.08% (P35)，流动性良好
- PIN 0.12，知情交易概率低
- 无明显盘口异常
- 建议：可正常下单，冲击成本可控参考 references/hft-microstructure-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 | 100,937 | 38,130 | -62% | 1 | 1 | 0% | 5,669 | 6,295 | +11% | 0 | 0 | — |
case-02 | fail→fail | 10,068 | 10,798 | +7% | 1 | 1 | 0% | 1,417 | 1,553 | +10% | 0 | 0 | — |
case-03 | fail→fail | 37,015 | 48,058 | +30% | 1 | 1 | 0% | 5,256 | 9,147 | +74% | 0 | 0 | — |
case-04 | pass→pass | 12,273 | 14,385 | +17% | 1 | 1 | 0% | 2,054 | 3,645 | +77% | 0 | 0 | — |
case-05 | pass→pass | 18,897 | 16,805 | -11% | 1 | 1 | 0% | 3,291 | 3,839 | +17% | 0 | 0 | — |
case-06 | pass→pass | 24,542 | 23,248 | -5% | 1 | 1 | 0% | 3,376 | 4,242 | +26% | 0 | 0 | — |
case-07 | pass→pass | 22,845 | 27,462 | +20% | 1 | 1 | 0% | 3,239 | 4,413 | +36% | 0 | 0 | — |
case-08 | fail→pass | 22,308 | 25,997 | +17% | 1 | 1 | 0% | 2,948 | 4,401 | +49% | 0 | 0 | — |
case-09 | pass→pass | 64,745 | 25,952 | -60% | 1 | 1 | 0% | 3,438 | 4,234 | +23% | 0 | 0 | — |
case-10 | pass→pass | 32,796 | 27,140 | -17% | 1 | 1 | 0% | 4,253 | 4,983 | +17% | 0 | 0 | — |
case-11 | pass→pass | 24,329 | 25,981 | +7% | 1 | 1 | 0% | 3,544 | 4,578 | +29% | 0 | 0 | — |
case-12 | pass→pass | 13,135 | 16,898 | +29% | 1 | 1 | 0% | 1,734 | 3,504 | +102% | 0 | 0 | — |
case-13 | pass→pass | 16,746 | 20,735 | +24% | 1 | 1 | 0% | 2,446 | 3,528 | +44% | 0 | 0 | — |
case-14 | fail→pass | 27,308 | 68,593 | +151% | 1 | 1 | 0% | 3,492 | 5,087 | +46% | 0 | 0 | — |
case-15 | fail→fail | 22,656 | 20,242 | -11% | 1 | 1 | 0% | 2,762 | 3,115 | +13% | 0 | 0 | — |
case-21 | pass→pass | 19,508 | 10,618 | -46% | 1 | 1 | 0% | 3,439 | 3,268 | -5% | 0 | 0 | — |
case-16 | pass→pass | 19,176 | 19,727 | +3% | 1 | 1 | 0% | 3,026 | 3,313 | +9% | 0 | 0 | — |
case-17 | pass→pass | 15,418 | 10,423 | -32% | 1 | 1 | 0% | 2,230 | 2,508 | +12% | 0 | 0 | — |
case-18 | pass→pass | 20,866 | 22,775 | +9% | 1 | 1 | 0% | 2,627 | 4,279 | +63% | 0 | 0 | — |
case-19 | pass→pass | 14,039 | 4,560 | -68% | 1 | 1 | 0% | 2,053 | 1,459 | -29% | 0 | 0 | — |
case-20 | pass→pass | 19,998 | 27,040 | +35% | 1 | 1 | 0% | 3,423 | 5,637 | +65% | 0 | 0 | — |
case-22 | pass→pass | 26,812 | 26,031 | -3% | 1 | 1 | 0% | 3,765 | 4,891 | +30% | 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, and 21 counted toward the lift figure. The other 1 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 +9 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.