Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股可解释AI/模型归因分析。当用户说"可解释AI"、"XAI"、"SHAP"、"模型解释"、"模型归因"、"explainable"、"为什么模型这样预测"时触发。基于 cn-stock-data 获取数据，对量化模型进行可解释性分析。支持 formal/brief 两种输出风格。
.claude/skills/aifinlab-a-share-explainable-ai/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 21% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 29% | 0% |
通过 cn-stock-data skill 获取数据：
# 可解释AI分析报告
## 一、全局解释
| 特征 | 重要性 | 方向 |
|------|--------|------|
## 二、局部解释
[个股SHAP瀑布图]
## 三、模型诊断
[偏差分析、漂移检测]
## 四、优化建议
[基于解释的改进方向]## XAI速览
- Top3特征：换手率/20日动量/ROE
- 个股解释：XX被选中因为动量强+质量高
- 模型无明显偏差
- 建议：去除2个负贡献特征参考 references/explainable-ai-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 | 42,984 | 52,506 | +22% | 1 | 1 | 0% | 6,069 | 6,911 | +14% | 0 | 0 | — |
case-02 | fail→pass | 19,501 | 21,311 | +9% | 1 | 1 | 0% | 2,928 | 2,894 | -1% | 0 | 0 | — |
case-03 | fail→fail | 30,174 | 46,857 | +55% | 1 | 1 | 0% | 4,648 | 8,344 | +80% | 0 | 0 | — |
case-04 | fail→pass | 23,248 | 25,948 | +12% | 1 | 1 | 0% | 3,554 | 4,801 | +35% | 0 | 0 | — |
case-05 | pass→pass | 20,151 | 23,590 | +17% | 1 | 1 | 0% | 3,130 | 4,043 | +29% | 0 | 0 | — |
case-06 | pass→pass | 24,190 | 23,460 | -3% | 1 | 1 | 0% | 3,691 | 4,271 | +16% | 0 | 0 | — |
case-07 | pass→pass | 23,035 | 21,966 | -5% | 1 | 1 | 0% | 2,958 | 3,536 | +20% | 0 | 0 | — |
case-08 | pass→pass | 20,241 | 15,602 | -23% | 1 | 1 | 0% | 2,950 | 2,988 | +1% | 0 | 0 | — |
case-09 | pass→pass | 22,028 | 21,222 | -4% | 1 | 1 | 0% | 3,188 | 3,886 | +22% | 0 | 0 | — |
case-10 | pass→pass | 12,829 | 8,098 | -37% | 1 | 1 | 0% | 1,680 | 1,919 | +14% | 0 | 0 | — |
case-11 | pass→pass | 20,553 | 24,633 | +20% | 1 | 1 | 0% | 3,094 | 3,772 | +22% | 0 | 0 | — |
case-12 | fail→pass | 16,569 | 4,053 | -76% | 1 | 1 | 0% | 2,660 | 1,153 | -57% | 0 | 0 | — |
case-13 | pass→pass | 21,108 | 26,489 | +25% | 1 | 1 | 0% | 3,017 | 4,430 | +47% | 0 | 0 | — |
case-14 | fail→fail | 17,202 | 18,242 | +6% | 1 | 1 | 0% | 2,727 | 2,449 | -10% | 0 | 0 | — |
case-15 | fail→fail | 25,415 | 26,253 | +3% | 1 | 1 | 0% | 3,684 | 4,740 | +29% | 0 | 0 | — |
case-16 | fail→fail | 21,195 | 19,861 | -6% | 1 | 1 | 0% | 2,954 | 3,403 | +15% | 0 | 0 | — |
case-17 | fail→fail | 31,431 | 35,197 | +12% | 1 | 1 | 0% | 4,700 | 5,182 | +10% | 0 | 0 | — |
case-18 | fail→fail | 33,137 | 31,263 | -6% | 1 | 1 | 0% | 4,682 | 4,949 | +6% | 0 | 0 | — |
case-19 | fail→pass | 12,379 | 11,840 | -4% | 1 | 1 | 0% | 1,972 | 2,390 | +21% | 0 | 0 | — |
case-20 | fail→fail | 33,775 | 30,715 | -9% | 1 | 1 | 0% | 5,008 | 6,402 | +28% | 0 | 0 | — |
case-21 | fail→fail | 15,550 | 22,481 | +45% | 1 | 1 | 0% | 2,926 | 4,658 | +59% | 0 | 0 | — |
case-22 | fail→fail | 25,425 | 23,166 | -9% | 1 | 1 | 0% | 3,370 | 3,829 | +14% | 0 | 0 | — |
case-23 | pass→pass | 20,015 | 22,373 | +12% | 1 | 1 | 0% | 2,777 | 3,909 | +41% | 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. 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.