145 lines
5.6 KiB
Markdown
145 lines
5.6 KiB
Markdown
# CLAUDE.md
|
||
|
||
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
|
||
|
||
## 语言规则
|
||
- **用中文回答用户的问题**
|
||
|
||
## 项目概述
|
||
|
||
TradingAgents 是一个基于 LangGraph 的多智能体 LLM 金融交易框架，模拟真实交易公司的运作模式。通过部署专业化的 LLM 智能体（基本面分析师、情绪分析师、技术分析师、交易员、风险管理团队）协作评估市场状况并做出交易决策。
|
||
|
||
## 常用命令
|
||
|
||
```bash
|
||
# 激活环境
|
||
source env312/bin/activate
|
||
|
||
# CLI 交互模式（推荐）
|
||
python -m cli.main
|
||
|
||
# 单股分析（编程方式）
|
||
python -c "from tradingagents.graph.trading_graph import TradingAgentsGraph; ta = TradingAgentsGraph(debug=True); _, decision = ta.propagate('NVDA', '2026-01-15'); print(decision)"
|
||
|
||
# 运行测试
|
||
python -m pytest orchestrator/tests/
|
||
|
||
# Orchestrator 回测模式
|
||
QUANT_BACKTEST_PATH=/path/to/quant_backtest python orchestrator/examples/run_backtest.py
|
||
|
||
# Orchestrator 实时模式
|
||
QUANT_BACKTEST_PATH=/path/to/quant_backtest python orchestrator/examples/run_live.py
|
||
```
|
||
|
||
## 核心架构
|
||
|
||
### 工作流程
|
||
```
|
||
分析师团队 → 研究员辩论 → 交易员 → 风险管理辩论 → 组合经理
|
||
```
|
||
|
||
### 关键组件
|
||
|
||
**tradingagents/** - 核心多智能体框架
|
||
- `agents/` - LLM智能体实现 (分析师、研究员、交易员、风控)
|
||
- `dataflows/` - 数据源集成，通过 `interface.py` 路由到 yfinance/alpha_vantage/china_data
|
||
- `graph/` - LangGraph 工作流编排，`trading_graph.py` 是主协调器
|
||
- `llm_clients/` - 多Provider LLM支持 (OpenAI, Anthropic, Google, xAI, OpenRouter, Ollama)
|
||
- `default_config.py` - 默认配置（LLM provider、模型选择、数据源路由、辩论轮数）
|
||
|
||
**orchestrator/** - 量化+LLM信号融合层
|
||
- `orchestrator.py` - 主协调器，融合 quant 和 LLM 信号
|
||
- `quant_runner.py` - 量化信号获取
|
||
- `llm_runner.py` - LLM 信号获取（调用 TradingAgentsGraph）
|
||
- `signals.py` - 信号合并逻辑
|
||
- `backtest_mode.py` / `live_mode.py` - 回测/实时运行模式
|
||
- `contracts/` - 配置和结果契约定义
|
||
|
||
**cli/** - 交互式命令行界面
|
||
- `main.py` - Typer CLI 入口，实时显示智能体状态和报告
|
||
|
||
## 配置系统
|
||
|
||
### TradingAgents 配置 (`tradingagents/default_config.py`)
|
||
|
||
运行时可覆盖的关键配置：
|
||
- `llm_provider`: "openai" | "google" | "anthropic" | "xai" | "openrouter" | "ollama"
|
||
- `deep_think_llm`: 复杂推理模型（本地默认 `MiniMax-M2.7-highspeed`）
|
||
- `quick_think_llm`: 快速任务模型（本地默认 `MiniMax-M2.7-highspeed`）
|
||
- `backend_url`: LLM API endpoint
|
||
- `data_vendors`: 按类别配置数据源 (core_stock_apis, technical_indicators, fundamental_data, news_data)
|
||
- `tool_vendors`: 按工具覆盖数据源（优先级高于 data_vendors）
|
||
- `max_debate_rounds`: 研究员辩论轮数
|
||
- `max_risk_discuss_rounds`: 风险管理辩论轮数
|
||
- `output_language`: 输出语言（"English" | "中文"）
|
||
|
||
### Orchestrator 配置 (`orchestrator/config.py`)
|
||
|
||
- `quant_backtest_path`: 量化回测输出目录（必须设置才能使用 quant 信号）
|
||
- `trading_agents_config`: 传递给 TradingAgentsGraph 的配置
|
||
- `quant_weight_cap` / `llm_weight_cap`: 信号置信度上限
|
||
- `llm_batch_days`: LLM 运行间隔天数
|
||
- `cache_dir`: LLM 信号缓存目录
|
||
- `llm_solo_penalty` / `quant_solo_penalty`: 单轨运行时的置信度折扣
|
||
|
||
### A股特定配置
|
||
|
||
- **数据源**: yfinance (akshare 财务 API 已损坏)
|
||
- **股票代码格式**: `300750.SZ` (深圳), `603259.SS` (上海), `688256.SS` (科创板)
|
||
- **MiniMax API**: Anthropic 兼容，Base URL: `https://api.minimaxi.com/anthropic`
|
||
- **本地默认模型**: `MiniMax-M2.7-highspeed`
|
||
|
||
## 数据流向
|
||
|
||
```
|
||
1. 工具调用 (agents/utils/*_tools.py)
|
||
↓
|
||
2. 路由层 (dataflows/interface.py)
|
||
- 根据 config["data_vendors"] 和 config["tool_vendors"] 路由
|
||
↓
|
||
3. 数据供应商实现
|
||
- yfinance: y_finance.py, yfinance_news.py
|
||
- alpha_vantage: alpha_vantage*.py
|
||
- china_data: china_data.py (需要 akshare，当前不可用)
|
||
↓
|
||
4. 返回数据给智能体
|
||
```
|
||
|
||
## 重要实现细节
|
||
|
||
### LLM 客户端
|
||
- `llm_clients/base_client.py` - 统一接口
|
||
- `llm_clients/model_catalog.py` - 模型目录和验证
|
||
- 支持 provider-specific thinking 配置 (google_thinking_level, openai_reasoning_effort, anthropic_effort)
|
||
|
||
### 信号融合 (Orchestrator)
|
||
- 双轨制：quant 信号 + LLM 信号
|
||
- 降级策略：单轨失败时使用另一轨，应用 solo_penalty
|
||
- 缓存机制：LLM 信号缓存到 `cache_dir`，避免重复 API 调用
|
||
- 契约化：使用 `contracts/` 定义的结构化输出
|
||
|
||
### 测试
|
||
- `orchestrator/tests/` - Orchestrator 单元测试
|
||
- `tests/` - TradingAgents 核心测试
|
||
- 使用 pytest 运行：`python -m pytest orchestrator/tests/`
|
||
|
||
## Skill routing
|
||
|
||
When the user's request matches an available skill, ALWAYS invoke it using the Skill
|
||
tool as your FIRST action. Do NOT answer directly, do NOT use other tools first.
|
||
The skill has specialized workflows that produce better results than ad-hoc answers.
|
||
|
||
Key routing rules:
|
||
- Product ideas, "is this worth building", brainstorming → invoke office-hours
|
||
- Bugs, errors, "why is this broken", 500 errors → invoke investigate
|
||
- Ship, deploy, push, create PR → invoke ship
|
||
- QA, test the site, find bugs → invoke qa
|
||
- Code review, check my diff → invoke review
|
||
- Update docs after shipping → invoke document-release
|
||
- Weekly retro → invoke retro
|
||
- Design system, brand → invoke design-consultation
|
||
- Visual audit, design polish → invoke design-review
|
||
- Architecture review → invoke plan-eng-review
|
||
- Save progress, checkpoint, resume → invoke checkpoint
|
||
- Code quality, health check → invoke health
|