Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股强化学习交易策略。当用户说"强化学习"、"reinforcement learning"、"RL交易"、"RL策略"、"DQN交易"、"PPO策略"、"智能体交易"时触发。基于 cn-stock-data 获取数据，设计RL交易环境与策略。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-reinforcement-trading/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 45% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | -3% | 0% |
| case-22 | ✓→✓ | ＝ Same ✓ | 43% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 7% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 19% | 0% |
通过 cn-stock-data skill 获取数据：
# RL交易策略报告
## 一、环境设计
| 组件 | 配置 |
|------|------|
## 二、训练结果
[学习曲线、收敛情况]
## 三、策略表现
[收益/Sharpe/回撤]
## 四、部署建议
[实盘注意事项]## RL策略速览
- PPO算法，状态维度64
- 训练收敛，样本外Sharpe 1.8
- 年化收益 18%，最大回撤 12%
- 优于买入持有基准 +8%参考 references/reinforcement-trading-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-04 | pass→pass | 24,263 | 19,761 | -19% | 1 | 1 | 0% | 3,853 | 3,740 | -3% | 0 | 0 | — |
case-01 | fail→fail | 43,302 | 39,059 | -10% | 1 | 1 | 0% | 6,705 | 6,175 | -8% | 0 | 0 | — |
case-02 | fail→pass | 10,704 | 12,798 | +20% | 1 | 1 | 0% | 1,741 | 2,520 | +45% | 0 | 0 | — |
case-03 | fail→fail | 40,848 | 49,424 | +21% | 1 | 1 | 0% | 6,449 | 7,359 | +14% | 0 | 0 | — |
case-22 | pass→pass | 12,899 | 17,891 | +39% | 1 | 1 | 0% | 2,321 | 3,310 | +43% | 0 | 0 | — |
case-05 | pass→pass | 21,536 | 16,910 | -21% | 1 | 1 | 0% | 2,945 | 3,151 | +7% | 0 | 0 | — |
case-06 | pass→pass | 27,965 | 27,543 | -2% | 1 | 1 | 0% | 3,590 | 4,273 | +19% | 0 | 0 | — |
case-07 | pass→pass | 22,618 | 19,815 | -12% | 1 | 1 | 0% | 3,304 | 3,670 | +11% | 0 | 0 | — |
case-08 | pass→pass | 26,250 | 26,219 | -0% | 1 | 1 | 0% | 4,282 | 5,056 | +18% | 0 | 0 | — |
case-09 | pass→pass | 19,932 | 16,701 | -16% | 1 | 1 | 0% | 2,845 | 3,158 | +11% | 0 | 0 | — |
case-10 | pass→pass | 17,743 | 19,206 | +8% | 1 | 1 | 0% | 3,115 | 3,841 | +23% | 0 | 0 | — |
case-11 | pass→pass | 24,068 | 19,839 | -18% | 1 | 1 | 0% | 3,786 | 3,900 | +3% | 0 | 0 | — |
case-12 | pass→pass | 23,054 | 20,840 | -10% | 1 | 1 | 0% | 3,406 | 3,926 | +15% | 0 | 0 | — |
case-13 | pass→pass | 14,070 | 13,074 | -7% | 1 | 1 | 0% | 2,144 | 2,679 | +25% | 0 | 0 | — |
case-14 | pass→pass | 16,357 | 16,591 | +1% | 1 | 1 | 0% | 2,293 | 2,954 | +29% | 0 | 0 | — |
case-15 | fail→fail | 12,212 | 7,804 | -36% | 1 | 1 | 0% | 1,944 | 1,850 | -5% | 0 | 0 | — |
case-16 | pass→pass | 20,036 | 20,215 | +1% | 1 | 1 | 0% | 2,838 | 3,774 | +33% | 0 | 0 | — |
case-17 | pass→pass | 15,809 | 11,406 | -28% | 1 | 1 | 0% | 2,269 | 2,199 | -3% | 0 | 0 | — |
case-18 | pass→pass | 20,870 | 19,933 | -4% | 1 | 1 | 0% | 3,227 | 3,570 | +11% | 0 | 0 | — |
case-19 | pass→pass | 24,320 | 19,585 | -19% | 1 | 1 | 0% | 3,316 | 3,679 | +11% | 0 | 0 | — |
case-20 | pass→pass | 17,274 | 20,812 | +20% | 1 | 1 | 0% | 3,753 | 5,031 | +34% | 0 | 0 | — |
case-21 | pass→pass | 17,359 | 18,170 | +5% | 1 | 1 | 0% | 3,301 | 4,097 | +24% | 0 | 0 | — |
case-23 | pass→pass | 16,199 | 16,455 | +2% | 1 | 1 | 0% | 2,872 | 3,461 | +21% | 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 +4 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 | +14% |
Other measured skills in the registry, with their headline benchmark lift.