股票预测系统

基于多因子分析，每小时自动预测沪深股票走势，支持参数自优化，目标准确率达90%以上。

Security Scan

The skill's code largely matches a stock-prediction tool, but there are notable mismatches between the documented usage and the code (hardcoded absolute workspace paths and file I/O) that are disproportionate to the stated instructions and could lead to unexpected access to files on the host.

Purpose & Capability

Name/description promise hourly multi-factor predictions — code implements that and includes ML models and prediction logic (reasonable). However several modules (stock_predict_ml.py, run_sim.py, auto_verify.py) use hardcoded absolute paths under /home/admin/openclaw/workspace (predictions, temp, sim_trading); SKILL.md claims outputs go to the local data/ directory and instructs running scripts from repo. These absolute path usages are not documented and are disproportionate to the stated purpose because they assume and access an external workspace outside the repository.

Instruction Scope

SKILL.md instructs running python scripts in src/ and storing outputs under data/. The source code does network I/O to public finance APIs (expected) but also reads/writes to absolute host paths (/home/admin/openclaw/workspace/*) not mentioned in SKILL.md. auto_verify/run_sim try to load predictions, configs and save results under that workspace; this expands scope beyond simple local repo execution and is not documented.

Install Mechanism

No install spec in registry (instruction-only), so nothing is auto-downloaded by the platform. requirements.txt lists heavy ML libraries (torch, xgboost, lightgbm) which are plausible for the claimed ML features but are large and may require significant privileges to install. No remote code-download URLs or extract operations were found.

Credentials

The skill declares no environment variables or credentials (good), but the code's hardcoded absolute paths (/home/admin/openclaw/workspace, /home/admin/openclaw/workspace/predictions, /home/admin/openclaw/workspace/temp, etc.) give it access to the host workspace and other files unrelated to the repository. That file-system access is disproportionate because SKILL.md implies data is local to the project (data/). The skill does not request secrets but could read arbitrary files in those paths if present.

Persistence & Privilege

The skill is not marked always:true and does not request to modify other skills or global agent settings. It will run only when invoked. That said, its code writes to host workspace paths (creating/modifying JSON, CSV files), so installing/running it will persist data in those locations — this is expected for a simulation tool but should be noted.

What to consider before installing

This package implements a plausible stock-prediction system, but before installing or running it you should: 1) Inspect and, if necessary, modify hardcoded paths in src/ (search for /home/admin/openclaw/workspace) so the tool only reads/writes inside the project folder (data/, predictions/). 2) Run it in a sandbox or isolated environment (not on a machine with sensitive files under /home/admin) because the code reads/writes absolute workspace paths that could include unrelated data. 3) Verify network calls are acceptable (it queries public finance APIs) and ensure your environment allows those requests. 4) Be skeptical of the “90% accuracy” claim — evaluate model outputs on historical data yourself. 5) If you plan to use it operationally, pin and audit dependencies (torch, xgboost, lightgbm) before installing. If you cannot confirm or change the hardcoded workspace paths, treat the skill as higher risk and avoid running it on systems containing sensitive files.

SKILL.md - 股票预测系统

技能描述

基于多因子分析的股票走势预测系统，支持每小时自动预测和参数自优化。

功能特点

• 🧠 多数据源支持: 腾讯财经、新浪财经、东方财富
• 🚀 并发处理: 多线程并发获取数据，5 分钟完成 300+ 股票分析
• 📊 智能预测: 基于趋势评分、动量因子、技术指标的综合预测
• 🔄 自动优化: 每 2 小时自动验证 10 支股票，自动调整参数
• 🎯 目标准确率: 90%+

安装依赖

pip install -r requirements.txt

系统要求:

• Python 3.8+
• numpy, pandas, requests

使用方法

运行预测

python src/stock_predict_hourly.py

自动验证

python src/auto_verify.py

查看预测结果

预测结果保存在 data/ 目录下：

• predictions.json - 预测结果
• performance.json - 性能记录

配置参数

核心参数

参数	默认值	说明
趋势评分系数	0.8	趋势评分权重
收益率窗口	15 日	计算收益率的窗口
高置信阈值	1.5%	高置信度预测阈值
中置信阈值	0.8%	中置信度预测阈值
预测方向阈值	±0.8%	预测涨跌阈值

自定义配置

编辑 src/stock_predict_hourly.py 中的配置：

TREND_COEF = 0.8  # 趋势评分系数
RETURN_WINDOW = 15  # 收益率窗口
HIGH_CONF_THRESHOLD = 1.5  # 高置信阈值
MID_CONF_THRESHOLD = 0.8  # 中置信阈值
PREDICT_THRESHOLD = 0.8  # 预测方向阈值

性能指标

v12.0 性能

指标	数值	说明
高置信度预测	63.4%	+22.7% 提升
平均预测幅度	2.75%	+98% 提升
方向稳定性	80.0%	达标

文件说明

stock-predictor/
├── src/
│   ├── stock_predict_hourly.py  # 每小时预测脚本
│   └── auto_verify.py           # 自动验证脚本
├── sim_trading/                  # 模拟交易数据
├── data/                         # 预测数据
├── docs/                         # 文档
├── tests/                        # 测试
├── requirements.txt              # Python 依赖
└── README.md                     # 详细文档

输出格式

预测结果 JSON 格式：

{
  "stock_code": "sz000001",
  "stock_name": "平安银行",
  "predict_change": 2.75,
  "confidence": "high",
  "trend_score": 8.5,
  "momentum": 0.03,
  "timestamp": "2026-03-16T22:00:00Z"
}

常见问题

Q: 预测准确率是多少？ A: v12.0 高置信度预测达到 63.4%，目标准确率 90%+

Q: 多久更新一次预测？ A: 每小时自动更新一次

Q: 支持哪些股票市场？ A: 支持 A 股（沪深两市）

Q: 如何查看历史预测？ A: 查看 data/predictions.json 文件

许可证

MIT License

作者

yun520-1

GitHub

https://github.com/yun520-1/stock-predictor