Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股ML股价预测/收益率预测。当用户说"ML预测"、"机器学习预测"、"股价预测"、"收益率预测"、"预测模型"、"ML选股"时触发。基于 cn-stock-data 获取数据,构建ML预测模型。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-ml-stock-predict/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-23 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 11% | 0% |
通过 cn-stock-data skill 获取数据:
# ML收益率预测报告
## 一、模型概览
| 模型 | 特征数 | 训练期 |
|------|--------|--------|
## 二、预测表现
[IC/ICIR/多空收益]
## 三、当期预测
[Top/Bottom股票列表]
## 四、模型健康度
[漂移检测、信号衰减]## ML预测速览
- LightGBM模型,128个特征
- 样本外IC=0.04, ICIR=1.5
- 本期Top10预测:[股票列表]
- 模型健康:信号稳定,无漂移参考 references/ml-stock-predict-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 | 37,107 | 34,116 | -8% | 1 | 1 | 0% | 5,403 | 5,205 | -4% | 0 | 0 | — |
case-02 | fail→pass | 11,505 | 7,438 | -35% | 1 | 1 | 0% | 2,033 | 2,011 | -1% | 0 | 0 | — |
case-03 | fail→fail | 30,966 | 27,876 | -10% | 1 | 1 | 0% | 4,921 | 5,321 | +8% | 0 | 0 | — |
case-04 | pass→pass | 24,524 | 22,406 | -9% | 1 | 1 | 0% | 3,788 | 4,212 | +11% | 0 | 0 | — |
case-05 | pass→pass | 21,972 | 19,998 | -9% | 1 | 1 | 0% | 3,231 | 3,612 | +12% | 0 | 0 | — |
case-06 | pass→pass | 23,655 | 26,981 | +14% | 1 | 1 | 0% | 3,818 | 4,241 | +11% | 0 | 0 | — |
case-07 | pass→pass | 20,813 | 20,270 | -3% | 1 | 1 | 0% | 3,256 | 3,824 | +17% | 0 | 0 | — |
case-08 | pass→pass | 22,461 | 23,138 | +3% | 1 | 1 | 0% | 3,304 | 3,637 | +10% | 0 | 0 | — |
case-09 | pass→pass | 20,986 | 20,909 | -0% | 1 | 1 | 0% | 2,973 | 3,885 | +31% | 0 | 0 | — |
case-10 | pass→pass | 19,527 | 18,360 | -6% | 1 | 1 | 0% | 2,983 | 3,642 | +22% | 0 | 0 | — |
case-11 | pass→pass | 21,719 | 20,290 | -7% | 1 | 1 | 0% | 3,273 | 3,881 | +19% | 0 | 0 | — |
case-12 | pass→pass | 26,222 | 24,195 | -8% | 1 | 1 | 0% | 3,828 | 4,378 | +14% | 0 | 0 | — |
case-13 | pass→pass | 25,237 | 28,619 | +13% | 1 | 1 | 0% | 3,456 | 4,596 | +33% | 0 | 0 | — |
case-14 | pass→pass | 20,575 | 13,328 | -35% | 1 | 1 | 0% | 2,764 | 2,733 | -1% | 0 | 0 | — |
case-15 | pass→pass | 23,126 | 24,444 | +6% | 1 | 1 | 0% | 3,440 | 3,826 | +11% | 0 | 0 | — |
case-16 | pass→pass | 25,484 | 23,581 | -7% | 1 | 1 | 0% | 3,485 | 4,477 | +28% | 0 | 0 | — |
case-17 | fail→pass | 19,180 | 17,587 | -8% | 1 | 1 | 0% | 2,984 | 3,443 | +15% | 0 | 0 | — |
case-18 | fail→pass | 12,597 | 2,351 | -81% | 1 | 1 | 0% | 1,872 | 1,051 | -44% | 0 | 0 | — |
case-19 | fail→fail | 16,681 | 3,865 | -77% | 1 | 1 | 0% | 2,563 | 1,225 | -52% | 0 | 0 | — |
case-20 | fail→fail | 34,614 | 31,890 | -8% | 1 | 1 | 0% | 6,351 | 6,896 | +9% | 0 | 0 | — |
case-21 | fail→fail | 27,166 | 56,956 | +110% | 1 | 1 | 0% | 3,924 | 8,860 | +126% | 0 | 0 | — |
case-22 | fail→fail | 29,133 | 37,868 | +30% | 1 | 1 | 0% | 4,531 | 7,380 | +63% | 0 | 0 | — |
case-23 | fail→pass | 14,470 | 8,023 | -45% | 1 | 1 | 0% | 2,027 | 1,960 | -3% | 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 +17 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 | +20% |
Other measured skills in the registry, with their headline benchmark lift.