Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股LSTM时序预测/序列建模。当用户说"LSTM"、"时序预测"、"序列模型"、"RNN预测"、"GRU"、"循环神经网络"、"LSTM预测股价"时触发。基于 cn-stock-data 获取数据，构建LSTM时序预测模型。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-lstm-forecast/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-22 | ✓→✗ | ▼ Worse | -68% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 29% | 0% |
通过 cn-stock-data skill 获取数据：
# LSTM时序预测报告
## 一、模型配置
| 参数 | 设置 |
|------|------|
## 二、预测表现
[MSE/方向准确率/IC]
## 三、当期预测
[未来N日预测值与置信区间]
## 四、模型诊断
[残差分析、过拟合检查]## LSTM预测速览
- 2层LSTM，隐藏128，窗口30天
- 方向准确率 56%，IC=0.03
- 预测明日：涨概率 62%
- 5日预测区间：[24.8, 26.2]参考 references/lstm-forecast-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 | 41,077 | 31,907 | -22% | 1 | 1 | 0% | 7,817 | 5,612 | -28% | 0 | 0 | — |
case-02 | pass→pass | 12,366 | 17,190 | +39% | 1 | 1 | 0% | 2,197 | 2,827 | +29% | 0 | 0 | — |
case-03 | fail→pass | 40,100 | 38,953 | -3% | 1 | 1 | 0% | 6,195 | 7,339 | +18% | 0 | 0 | — |
case-04 | pass→pass | 19,060 | 18,340 | -4% | 1 | 1 | 0% | 2,949 | 3,660 | +24% | 0 | 0 | — |
case-05 | pass→pass | 16,806 | 19,389 | +15% | 1 | 1 | 0% | 2,713 | 3,341 | +23% | 0 | 0 | — |
case-06 | pass→pass | 23,511 | 21,607 | -8% | 1 | 1 | 0% | 2,931 | 3,513 | +20% | 0 | 0 | — |
case-07 | fail→pass | 18,178 | 17,819 | -2% | 1 | 1 | 0% | 3,023 | 3,641 | +20% | 0 | 0 | — |
case-08 | pass→pass | 21,725 | 19,611 | -10% | 1 | 1 | 0% | 3,257 | 3,394 | +4% | 0 | 0 | — |
case-09 | fail→fail | 17,075 | 19,492 | +14% | 1 | 1 | 0% | 2,723 | 3,838 | +41% | 0 | 0 | — |
case-10 | pass→pass | 18,207 | 18,860 | +4% | 1 | 1 | 0% | 2,681 | 3,721 | +39% | 0 | 0 | — |
case-11 | pass→pass | 16,768 | 17,446 | +4% | 1 | 1 | 0% | 2,679 | 3,144 | +17% | 0 | 0 | — |
case-12 | pass→pass | 22,899 | 20,977 | -8% | 1 | 1 | 0% | 3,232 | 4,218 | +31% | 0 | 0 | — |
case-13 | fail→pass | 21,202 | 8,343 | -61% | 1 | 1 | 0% | 3,378 | 2,020 | -40% | 0 | 0 | — |
case-14 | pass→pass | 14,084 | 12,148 | -14% | 1 | 1 | 0% | 1,911 | 2,571 | +35% | 0 | 0 | — |
case-15 | pass→pass | 15,693 | 18,828 | +20% | 1 | 1 | 0% | 2,634 | 3,788 | +44% | 0 | 0 | — |
case-16 | pass→pass | 18,304 | 17,015 | -7% | 1 | 1 | 0% | 2,746 | 3,168 | +15% | 0 | 0 | — |
case-17 | pass→pass | 19,666 | 24,481 | +24% | 1 | 1 | 0% | 2,983 | 3,938 | +32% | 0 | 0 | — |
case-18 | pass→pass | 22,580 | 21,109 | -7% | 1 | 1 | 0% | 3,349 | 3,839 | +15% | 0 | 0 | — |
case-19 | pass→pass | 19,315 | 18,694 | -3% | 1 | 1 | 0% | 2,874 | 3,601 | +25% | 0 | 0 | — |
case-20 | pass→pass | 19,568 | 22,862 | +17% | 1 | 1 | 0% | 2,998 | 3,731 | +24% | 0 | 0 | — |
case-21 | pass→pass | 21,381 | 23,737 | +11% | 1 | 1 | 0% | 3,205 | 4,082 | +27% | 0 | 0 | — |
case-22 | pass→fail | 28,824 | 9,680 | -66% | 1 | 1 | 0% | 4,329 | 1,368 | -68% | 0 | 0 | — |
case-23 | pass→pass | 25,339 | 33,341 | +32% | 1 | 1 | 0% | 4,679 | 6,870 | +47% | 0 | 0 | — |
case-24 | pass→pass | 27,225 | 41,293 | +52% | 1 | 1 | 0% | 4,044 | 6,174 | +53% | 0 | 0 | — |
case-25 | pass→pass | 28,665 | 24,220 | -16% | 1 | 1 | 0% | 4,506 | 5,407 | +20% | 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. 25 cases were attempted, and 24 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +8 percentage points is the difference between those two pass rates over the 24 comparable cases. 2 cases got worse with the skill loaded, and they are 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.