Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股Transformer量化/注意力机制因子。当用户说"Transformer"、"注意力机制"、"attention"、"自注意力"、"Transformer量化"、"GPT选股"时触发。基于 cn-stock-data 获取数据，构建Transformer量化模型。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-transformer-quant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | 16% | 0% |
通过 cn-stock-data skill 获取数据：
# Transformer量化报告
## 一、模型架构
| 组件 | 配置 |
|------|------|
## 二、注意力分析
[注意力权重可视化、关键时间步]
## 三、信号表现
[IC/多空收益/与传统模型对比]
## 四、当期信号
[Top/Bottom股票列表]## Transformer量化速览
- 6层Transformer，8头注意力
- IC=0.055，优于LightGBM(0.04)
- 注意力集中在近5日量价变化
- 本期Top信号：[股票列表]参考 references/transformer-quant-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 | 39,491 | 37,771 | -4% | 1 | 1 | 0% | 5,408 | 6,216 | +15% | 0 | 0 | — |
case-02 | fail→pass | 20,787 | 10,737 | -48% | 1 | 1 | 0% | 3,124 | 2,198 | -30% | 0 | 0 | — |
case-03 | fail→fail | 36,562 | 33,584 | -8% | 1 | 1 | 0% | 5,901 | 6,419 | +9% | 0 | 0 | — |
case-04 | fail→fail | 34,702 | 29,474 | -15% | 1 | 1 | 0% | 4,950 | 4,676 | -6% | 0 | 0 | — |
case-05 | fail→fail | 30,566 | 22,985 | -25% | 1 | 1 | 0% | 4,289 | 4,242 | -1% | 0 | 0 | — |
case-06 | fail→fail | 22,241 | 18,306 | -18% | 1 | 1 | 0% | 3,861 | 4,099 | +6% | 0 | 0 | — |
case-07 | pass→pass | 27,395 | 23,212 | -15% | 1 | 1 | 0% | 3,597 | 4,179 | +16% | 0 | 0 | — |
case-08 | pass→pass | 24,917 | 23,083 | -7% | 1 | 1 | 0% | 3,890 | 4,358 | +12% | 0 | 0 | — |
case-09 | pass→pass | 26,269 | 22,306 | -15% | 1 | 1 | 0% | 3,410 | 4,132 | +21% | 0 | 0 | — |
case-10 | pass→pass | 17,910 | 19,650 | +10% | 1 | 1 | 0% | 2,314 | 3,275 | +42% | 0 | 0 | — |
case-11 | pass→pass | 26,449 | 24,031 | -9% | 1 | 1 | 0% | 3,669 | 4,171 | +14% | 0 | 0 | — |
case-12 | fail→fail | 12,976 | 3,734 | -71% | 1 | 1 | 0% | 2,011 | 1,101 | -45% | 0 | 0 | — |
case-18 | pass→pass | 28,927 | 20,378 | -30% | 1 | 1 | 0% | 3,196 | 3,658 | +14% | 0 | 0 | — |
case-13 | fail→pass | 29,457 | 19,745 | -33% | 1 | 1 | 0% | 4,739 | 3,531 | -25% | 0 | 0 | — |
case-14 | fail→fail | 19,086 | 2,728 | -86% | 1 | 1 | 0% | 2,837 | 1,055 | -63% | 0 | 0 | — |
case-15 | fail→pass | 17,992 | 8,539 | -53% | 1 | 1 | 0% | 2,658 | 1,760 | -34% | 0 | 0 | — |
case-16 | pass→pass | 18,310 | 13,349 | -27% | 1 | 1 | 0% | 2,739 | 2,674 | -2% | 0 | 0 | — |
case-17 | pass→pass | 25,380 | 24,596 | -3% | 1 | 1 | 0% | 3,117 | 3,745 | +20% | 0 | 0 | — |
case-19 | pass→pass | 20,898 | 7,054 | -66% | 1 | 1 | 0% | 2,816 | 1,567 | -44% | 0 | 0 | — |
case-20 | pass→pass | 28,760 | 19,701 | -31% | 1 | 1 | 0% | 2,762 | 3,544 | +28% | 0 | 0 | — |
case-21 | fail→pass | 15,880 | 10,467 | -34% | 1 | 1 | 0% | 2,535 | 2,319 | -9% | 0 | 0 | — |
case-22 | fail→fail | 14,401 | 17,530 | +22% | 1 | 1 | 0% | 3,158 | 3,866 | +22% | 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 +18 percentage points is the difference between those two pass rates over the 22 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.