Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股XGBoost选股/梯度提升模型。当用户说"XGBoost"、"GBDT"、"梯度提升"、"LightGBM选股"、"树模型选股"、"XGB"、"机器学习选股"时触发。基于 cn-stock-data 获取数据，构建XGBoost/LightGBM选股模型。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-xgboost-screen/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 13% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 38% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 42% | 0% |
通过 cn-stock-data skill 获取数据：
# XGBoost选股报告
## 一、模型概览
| 参数 | 设置 |
|------|------|
## 二、特征重要性
[Top10特征及SHAP值]
## 三、选股表现
[IC/多空收益/分组单调性]
## 四、本期选股
[Top股票列表与预测分数]## XGBoost选股速览
- LightGBM，85个特征
- IC=0.042, ICIR=1.6
- Top特征：20日动量、ROE变化、换手率
- 本期Top10：[股票列表]参考 references/xgboost-screen-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-07 | fail→fail | 19,448 | 18,703 | -4% | 1 | 1 | 0% | 2,770 | 3,848 | +39% | 0 | 0 | — |
case-01 | fail→fail | 30,252 | 39,293 | +30% | 1 | 1 | 0% | 5,227 | 7,408 | +42% | 0 | 0 | — |
case-02 | fail→fail | 16,234 | 18,107 | +12% | 1 | 1 | 0% | 2,741 | 3,540 | +29% | 0 | 0 | — |
case-03 | fail→fail | 37,424 | 32,074 | -14% | 1 | 1 | 0% | 5,862 | 5,388 | -8% | 0 | 0 | — |
case-04 | pass→pass | 18,113 | 14,323 | -21% | 1 | 1 | 0% | 2,632 | 2,961 | +13% | 0 | 0 | — |
case-05 | pass→pass | 23,047 | 24,877 | +8% | 1 | 1 | 0% | 3,416 | 4,711 | +38% | 0 | 0 | — |
case-06 | pass→pass | 18,410 | 18,055 | -2% | 1 | 1 | 0% | 2,516 | 3,569 | +42% | 0 | 0 | — |
case-08 | pass→pass | 9,021 | 10,086 | +12% | 1 | 1 | 0% | 1,360 | 1,990 | +46% | 0 | 0 | — |
case-09 | pass→pass | 18,352 | 20,502 | +12% | 1 | 1 | 0% | 2,798 | 3,984 | +42% | 0 | 0 | — |
case-10 | pass→pass | 24,118 | 20,202 | -16% | 1 | 1 | 0% | 3,478 | 3,924 | +13% | 0 | 0 | — |
case-11 | fail→fail | 25,926 | 29,640 | +14% | 1 | 1 | 0% | 4,156 | 5,080 | +22% | 0 | 0 | — |
case-12 | fail→pass | 13,426 | 3,706 | -72% | 1 | 1 | 0% | 2,100 | 1,377 | -34% | 0 | 0 | — |
case-13 | pass→pass | 23,142 | 17,789 | -23% | 1 | 1 | 0% | 3,276 | 3,707 | +13% | 0 | 0 | — |
case-14 | pass→pass | 16,466 | 16,971 | +3% | 1 | 1 | 0% | 2,706 | 2,996 | +11% | 0 | 0 | — |
case-15 | pass→pass | 15,764 | 22,303 | +41% | 1 | 1 | 0% | 2,473 | 3,589 | +45% | 0 | 0 | — |
case-16 | fail→pass | 20,646 | 18,610 | -10% | 1 | 1 | 0% | 2,712 | 3,125 | +15% | 0 | 0 | — |
case-17 | pass→pass | 22,187 | 18,493 | -17% | 1 | 1 | 0% | 3,216 | 3,793 | +18% | 0 | 0 | — |
case-18 | pass→pass | 26,974 | 43,474 | +61% | 1 | 1 | 0% | 3,321 | 4,546 | +37% | 0 | 0 | — |
case-19 | pass→pass | 17,059 | 21,118 | +24% | 1 | 1 | 0% | 2,667 | 3,478 | +30% | 0 | 0 | — |
case-20 | fail→fail | 26,114 | 23,961 | -8% | 1 | 1 | 0% | 3,977 | 4,962 | +25% | 0 | 0 | — |
case-21 | fail→fail | 30,533 | 29,036 | -5% | 1 | 1 | 0% | 4,195 | 4,817 | +15% | 0 | 0 | — |
case-22 | fail→fail | 33,864 | 32,210 | -5% | 1 | 1 | 0% | 6,469 | 7,039 | +9% | 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. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 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.