Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股深度学习Alpha信号/神经网络量化。当用户说"深度学习"、"deep learning"、"神经网络量化"、"DL alpha"、"CNN选股"、"深度因子"时触发。基于 cn-stock-data 获取数据,构建深度学习Alpha模型。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-deep-learning-alpha/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 2% | 0% |
| case-01 | ✓→✓ | = Same ✓ | 4% | 0% |
通过 cn-stock-data skill 获取数据:
# 深度学习Alpha信号报告
## 一、模型架构
| 组件 | 配置 |
|------|------|
## 二、信号表现
[IC/多空收益/Sharpe]
## 三、当期Alpha信号
[Top/Bottom股票与信号强度]
## 四、归因分析
[特征贡献度、注意力权重]## DL Alpha速览
- Transformer模型,IC=0.05
- 多空年化 22%,Sharpe 2.1
- 本期强Alpha:[股票列表]
- 关键驱动:量价模式+资金流参考 references/deep-learning-alpha-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-05 | pass→pass | 20,419 | 18,628 | -9% | 1 | 1 | 0% | 3,664 | 3,732 | +2% | 0 | 0 | — |
case-01 | pass→pass | 49,592 | 55,571 | +12% | 1 | 1 | 0% | 7,760 | 8,103 | +4% | 0 | 0 | — |
case-02 | fail→fail | 17,567 | 25,843 | +47% | 1 | 1 | 0% | 2,832 | 4,498 | +59% | 0 | 0 | — |
case-03 | pass→pass | 51,069 | 39,007 | -24% | 1 | 1 | 0% | 8,244 | 6,867 | -17% | 0 | 0 | — |
case-04 | pass→pass | 14,891 | 12,274 | -18% | 1 | 1 | 0% | 2,599 | 2,429 | -7% | 0 | 0 | — |
case-06 | pass→pass | 20,120 | 20,842 | +4% | 1 | 1 | 0% | 3,146 | 4,058 | +29% | 0 | 0 | — |
case-07 | pass→pass | 20,375 | 21,197 | +4% | 1 | 1 | 0% | 3,061 | 3,595 | +17% | 0 | 0 | — |
case-08 | fail→pass | 15,192 | 18,130 | +19% | 1 | 1 | 0% | 2,285 | 2,987 | +31% | 0 | 0 | — |
case-09 | pass→pass | 19,802 | 17,738 | -10% | 1 | 1 | 0% | 3,077 | 3,334 | +8% | 0 | 0 | — |
case-20 | pass→pass | 14,435 | 2,717 | -81% | 1 | 1 | 0% | 1,985 | 1,060 | -47% | 0 | 0 | — |
case-10 | pass→pass | 27,411 | 35,500 | +30% | 1 | 1 | 0% | 4,435 | 5,824 | +31% | 0 | 0 | — |
case-11 | fail→fail | 15,444 | 4,649 | -70% | 1 | 1 | 0% | 2,378 | 1,345 | -43% | 0 | 0 | — |
case-12 | pass→pass | 19,758 | 19,894 | +1% | 1 | 1 | 0% | 3,115 | 3,986 | +28% | 0 | 0 | — |
case-13 | fail→fail | 11,114 | 10,107 | -9% | 1 | 1 | 0% | 1,867 | 2,259 | +21% | 0 | 0 | — |
case-14 | fail→pass | 19,157 | 19,327 | +1% | 1 | 1 | 0% | 2,806 | 3,848 | +37% | 0 | 0 | — |
case-15 | pass→pass | 22,810 | 24,053 | +5% | 1 | 1 | 0% | 3,330 | 4,672 | +40% | 0 | 0 | — |
case-16 | pass→pass | 21,365 | 21,308 | -0% | 1 | 1 | 0% | 3,065 | 3,811 | +24% | 0 | 0 | — |
case-17 | pass→pass | 19,170 | 19,671 | +3% | 1 | 1 | 0% | 3,146 | 3,469 | +10% | 0 | 0 | — |
case-18 | fail→pass | 27,227 | 24,987 | -8% | 1 | 1 | 0% | 3,887 | 5,082 | +31% | 0 | 0 | — |
case-19 | pass→pass | 13,048 | 10,083 | -23% | 1 | 1 | 0% | 2,011 | 2,193 | +9% | 0 | 0 | — |
case-21 | pass→pass | 30,756 | 49,634 | +61% | 1 | 1 | 0% | 5,013 | 8,702 | +74% | 0 | 0 | — |
case-22 | pass→pass | 22,671 | 24,759 | +9% | 1 | 1 | 0% | 2,929 | 4,895 | +67% | 0 | 0 | — |
case-23 | pass→pass | 19,360 | 22,997 | +19% | 1 | 1 | 0% | 2,969 | 3,864 | +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. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/28/2026 | +5% |
Other measured skills in the registry, with their headline benchmark lift.