功能描述

基于机器学习的双色球预测工具，随机森林+Gradient Boosting+ 规则集成，15 维特征工程，红球准确率 80.8%

使用说明 (SKILL.md)

🎰 彩票预测技能（V3.8 机器学习增强版）

Name: Lottery Predictor V3 8
Author: mumuli2021

基于 3431 期历史数据，采用机器学习 + 规则引擎混合集成系统，提供科学的双色球预测推荐。

红球预测准确率：80.8%（2000 期训练）

核心算法

机器学习架构

红球预测：
┌─────────────────────────────────────┐
│  RandomForest Classifier             │
│  - 100 棵树，最大深度 12 层            │
│  - 15 维特征工程                     │
│  - StandardScaler 标准化             │
│  - class_weight=balanced             │
└─────────────────────────────────────┘
              ↓
         红球概率输出

蓝球预测：
┌─────────────────────────────────────┐
│  Gradient Boosting Classifier        │
│  - 30 棵树，学习率 0.1                │
│  - 5 维特征                          │
└─────────────────────────────────────┘
              ↓
         蓝球概率输出

最终预测：
┌─────────────────────────────────────┐
│  混合集成 (ML 50% + 规则 50%)        │
└─────────────────────────────────────┘

15 维特征工程

特征类别	特征名	说明
基础统计	recent_10	近 10 期出现次数
	recent_30	近 30 期出现次数
	recent_50	近 50 期出现次数
遗漏特征	current_miss	当前遗漏值
	avg_miss	历史平均遗漏
	max_miss	历史最大遗漏
形态特征	is_repeat	是否重复号
	has_consecutive	连号标记
	same_tail_count	同尾号数量
频率特征	freq	历史出现频率
区间分布	in_zone1/2/3	区间标记
号码属性	is_odd	奇偶性
	is_big	大小号
	is_prime	质合数

训练配置

参数	值	说明
训练窗口	2000 期	最近 2000 期数据
训练样本	62,700	红球样本量
树数量	100 棵	随机森林
最大深度	12 层	防止过拟合
ML 权重	50%	与规则引擎集成

使用方法

基础预测

预测双色球下一期

用 lottery-predictor-v3-8 预测

查看回测

运行 V3.8 回测，1000 期

查看模型信息

显示 V3.8 模型配置

输出示例

================================================================================
🔮 V3.8 预测结果
================================================================================

上期开奖：01 03 07 13 22 23 + 07

【红球推荐】06, 07, 09, 11, 16, 19
【蓝球推荐】02

💡 说明：
  - 使用随机森林预测红球
  - 使用 Gradient Boosting 预测蓝球
  - 集成规则引擎提高稳定性
  - 训练数据：最近 2000 期

================================================================================
⚠️  预测仅供娱乐参考，请理性购彩！
================================================================================

性能对比

版本	算法	训练数据	红球准确率	蓝球准确率
V3.6	规则引擎	全部历史	77.2%(≥3 红)	-
V3.7	规则增强	全部历史	77.2%(≥3 红)	-
V3.8	ML+ 规则	2000 期	80.8%	93.8%

依赖安装

pip3 install scikit-learn numpy pandas

文件结构

lottery-predictor-v3-8/
├── SKILL.md              # 技能定义
├── package.json          # 包信息
├── README.md             # 使用说明
├── scripts/
│   ├── v3.8_ml_model.py  # 主程序
│   └── backtest_v3.8.py  # 回测脚本
├── assets/               # 资源文件
└── references/           # 参考资料

⚠️ 免责声明

本技能仅供娱乐参考，不保证中奖！

彩票本质是随机事件，历史规律不代表未来结果！

请理性购彩，量力而行！

版本：3.8.0 | 作者：田七七 💻 | 2026-03-30

安全使用建议

This skill's stated purpose (lottery prediction) matches the included ML scripts, but several inconsistencies make it suspicious rather than clearly benign. Before installing or running it, consider: 1) The skill metadata requires LOTTERY_DB_PATH, but both Python scripts ignore that env var and instead use a hard-coded database path (~/.openclaw/workspace/projects/caipiao/data/caipiao.db). Confirm which DB will actually be read and whether that path contains only non-sensitive data. 2) The package claims filesystem 'read' only, yet the code writes report files under ~/.openclaw/.../reports/ — expect write access. 3) package.json references a verify.py script that is not included; the repository may be incomplete or broken. 4) No network calls are present in the code (good), but the code will create/modify files in your home directory. Recommendations: run the code in an isolated/sandbox environment (or inspect & modify the DB_PATH constant to point to a safe test DB) and verify the full, un-truncated scripts before giving it access to real data. If you can, ask the publisher to fix the env var/metadata mismatch, provide a configurable DB path that respects LOTTERY_DB_PATH, and confirm intended filesystem writes. If you cannot verify these, avoid installing on a primary machine.

功能分析

Type: OpenClaw Skill Name: lottery-predictor-v3-8 Version: 3.8.0 The skill bundle implements a lottery prediction system using standard machine learning techniques (Random Forest and Gradient Boosting) via scikit-learn. The scripts (v3.8_ml_model.py and backtest_v3.8.py) perform local data processing on a historical SQLite database and generate predictions or backtest reports within the user's workspace. There is no evidence of data exfiltration, unauthorized network activity, or malicious code execution; the functionality aligns strictly with the stated purpose of the tool.

能力评估

⚠ Purpose & Capability

Name/description (lottery prediction) match required binaries (python3, sqlite3) and included ML code, so core purpose is plausible. However the skill declares LOTTERY_DB_PATH as a required environment variable / primary credential in metadata, while both scripts hard-code DB_PATH to '~/.openclaw/workspace/projects/caipiao/data/caipiao.db' and do not read LOTTERY_DB_PATH. Marking a filesystem path as a 'primary credential' is odd. package.json also lists 'env': ['LOTTERY_DB_PATH'], but the code never uses it — this is an incoherence between what the skill claims to need and what it actually uses.

⚠ Instruction Scope

SKILL.md instructs installing scikit-learn/numpy/pandas and running the included Python scripts; that matches the code. But the runtime instructions and metadata imply the agent will use LOTTERY_DB_PATH, while both scripts access a hard-coded path under the user's home and write reports to '~/.openclaw/workspace/projects/caipiao/reports/'. The skill's metadata does not declare these specific paths, and package.json claims filesystem 'read' permission only even though scripts write reports. The code also references other scripts (e.g., 'verify.py' in package.json) that are not present. These scope mismatches (expected env var vs hard-coded path, read-only permission vs write behavior) are concerning.

ℹ Install Mechanism

There is no automated install spec (instruction-only), and dependencies are standard PyPI packages (scikit-learn, numpy, pandas) recommended via pip. No external downloads or URL-based installers are present. This is low risk from an install-origin perspective, but installing ML packages is required to run the models.

⚠ Credentials

The only declared required env var is LOTTERY_DB_PATH (marked as primary credential), which is a path rather than a secret; that could be reasonable if the code used it. But the code ignores the env var and instead hard-codes DB_PATH to a specific location in the user's home directory. Declaring a path as a primary credential is misleading. Also package.json claims no network and only filesystem read, yet the scripts read and write files under ~/.openclaw/workspace/projects/caipiao/ (reads DB, writes reports) — an access mismatch that should be clarified before use.

⚠ Persistence & Privilege

The skill does not request 'always: true' and is user-invocable (normal). However the code writes reports to a persistent location under the user's home (~/.openclaw/.../reports/), while the package metadata claims read-only filesystem permission. The skill also assumes and uses a specific path under the user's home without declaring that path as a required config, which gives it persistent filesystem presence in an unexpected location. This combination of hard-coded paths and write behavior is a red flag for unexpected persistence.

版本历史

v3.8.0

- 增强版机器学习预测，红球准确率提升至 80.8% - 新增 15 维特征工程，融合基础统计、遗漏、形态等多维指标 - 红球采用随机森林（100 棵树，最大深度 12），蓝球引入 Gradient Boosting 集成 - 训练窗口改为近 2000 期，提升预测科学性 - 推理结果结合机器学习与规则引擎（各占 50% 权重） - 完善回测、模型信息等指令，文档内容更新完善

元数据

Slug lottery-predictor-v3-8

版本 3.8.0

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 1

常见问题

Lottery Predictor V3 8 是什么？

基于机器学习的双色球预测工具，随机森林+Gradient Boosting+ 规则集成，15 维特征工程，红球准确率 80.8%. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 105 次。

如何安装 Lottery Predictor V3 8？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install lottery-predictor-v3-8」即可一键安装，无需额外配置。

Lottery Predictor V3 8 是免费的吗？

是的，Lottery Predictor V3 8 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

Lottery Predictor V3 8 支持哪些平台？

Lottery Predictor V3 8 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 Lottery Predictor V3 8？

由 mumuli2021（@mumuli2021）开发并维护，当前版本 v3.8.0。

Lottery Predictor V3 8