investment-committee

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Investment Committee (Multi-Agent Adversarial V3)

启动一个由三位顶级投资人风格组成的虚拟委员会，对目标公司进行严格的独立审查、对抗式辩论和投票决议。

Launch a virtual committee modeled after three top investors to conduct rigorous independent reviews, adversarial debates, and voting resolutions on target companies.

⚔️ 核心设计原则

⚔️ Core Design Principles

物理隔离: 每个 Agent 是独立的 Gemini API 调用，拥有完整的 Persona 系统提示。
深度人设: 通用化投资哲学框架，不依赖特定案例。
数据注入: 自动抓取实时宏观数据（利率、汇率、VIX），并采用前向填充 (
```
ffill
```
) 处理节假日/缺失数据，确保德肯米勒始终获得有效情报。
智能重试: 内置指数退避 (Exponential Backoff) 机制，自动处理 API 429 限流错误。
决议机制: 辩论结束 --> 自动提取投票 --> 主席 Agent 形成决议。

Physical Isolation: Each Agent is an independent Gemini API call with a complete Persona system prompt.
In-Depth Persona: Generalized investment philosophy framework, not dependent on specific cases.
Data Injection: Automatically fetch real-time macro data (interest rates, exchange rates, VIX), and use forward fill (
```
ffill
```
) to handle holidays/missing data, ensuring Druckenmiller always receives valid intelligence.
Intelligent Retry: Built-in Exponential Backoff mechanism to automatically handle API 429 rate-limiting errors.
Resolution Mechanism: Debate ends --> automatically extract votes --> Chairman Agent forms resolution.

👥 委员会成员

👥 Committee Members

角色	投资哲学内核	数据注入
巴菲特	"生意"思维、护城河、安全边际	标准研报
木头姐	莱特定律、S曲线拐点、技术融合	标准研报
德肯米勒	流动性为王、价格行为、不对称赔率	实时宏观快照 (美债/美元/VIX)

详细人设见:

references/personas/

目录。

Role	Core Investment Philosophy	Data Injection
Buffett	"Business" mindset, moat, margin of safety	Standard research reports
Wood	Wright's Law, S-curve inflection point, technology integration	Standard research reports
Druckenmiller	Liquidity first, price action, asymmetric odds	Real-time macro snapshot (U.S. Treasuries/USD/VIX)

Detailed personas can be found in:

references/personas/

directory.

🔄 执行流程

🔄 Execution Process

Pre-Flight: 数据抓取

Pre-Flight: Data Fetching

脚本自动从 Yahoo Finance 抓取：

10年期美债收益率 (
```
^TNX
```
)
美元指数 (
```
DX-Y.NYB
```
)
恐慌指数 (
```
^VIX
```
)
标普/纳指趋势 (
```
SPY
```
,
```
QQQ
```
)

The script automatically fetches from Yahoo Finance:

10-year U.S. Treasury Yield (
```
^TNX
```
)
U.S. Dollar Index (
```
DX-Y.NYB
```
)
VIX (
```
^VIX
```
)
S&P/Nasdaq trends (
```
SPY
```
,
```
QQQ
```
)

Phase 1: 独立初评 (Independent Review)

Phase 1: Independent Review

每位专家阅读研报（德肯米勒额外获得宏观数据），给出独立判断。

Each expert reads the research report (Druckenmiller additionally receives macro data) and provides an independent judgment.

Phase 2: 对抗辩论 (Adversarial Debate)

Phase 2: Adversarial Debate

多轮辩论，针对性反驳，并更新立场。

Multi-round debates with targeted rebuttals and position updates.

Phase 3: 投票决议 (Voting & Decision)

Phase 3: Voting & Decision

提取每位专家的投票（买入/拒绝/观望）和置信度。
结合宏观背景生成《投资委员会最终决议》。

Extract each expert's vote (Buy/Reject/Hold) and confidence level.
Generate the Final Resolution of the Investment Committee combined with the macro background.

🛠️ 技术栈

🛠️ Tech Stack

LLM: Google Gemini 2.0 Flash (
```
google-genai
```
)
Data: Yahoo Finance (
```
yfinance
```
)
Env: 支持 HTTP 代理

LLM: Google Gemini 2.0 Flash (
```
google-genai
```
)
Data: Yahoo Finance (
```
yfinance
```
)
Env: Supports HTTP proxy

🚀 使用方法

🚀 Usage

1. 安装依赖

1. Install Dependencies

bash

pip install -r requirements.txt

bash

pip install -r requirements.txt

2. 设置环境

2. Set Up Environment

bash

undefined

bash

undefined

Windows PowerShell

$env:GEMINI_API_KEY='<YOUR_API_KEY>' # 替换为你的 Gemini API Key

$env:GEMINI_API_KEY='<YOUR_API_KEY>' # Replace with your Gemini API Key

可选代理（如使用 Clash/V2Ray 等）

Optional proxy (e.g., using Clash/V2Ray, etc.)

$env:HTTP_PROXY='http://127.0.0.1:<PORT>' # 替换为你的代理端口 $env:HTTPS_PROXY='http://127.0.0.1:<PORT>'


> **获取 API Key**: 访问 [Google AI Studio](https://aistudio.google.com/app/apikey) 创建 API Key

$env:HTTP_PROXY='http://127.0.0.1:<PORT>' # Replace with your proxy port $env:HTTPS_PROXY='http://127.0.0.1:<PORT>'


> **Get API Key**: Visit [Google AI Studio](https://aistudio.google.com/app/apikey) to create an API Key

3. 运行

3. Run

bash

python scripts/run_committee.py <path_to_report.md> --rounds 3 --output ./output

bash

python scripts/run_committee.py <path_to_report.md> --rounds 3 --output ./output

📁 Skill 目录结构

📁 Skill Directory Structure

investment-committee/
├── SKILL.md                      # 本文件 (架构说明)
├── TROUBLESHOOTING.md            # 问题排查
├── requirements.txt              # 依赖 (google-genai, yfinance)
├── scripts/
│   └── run_committee.py          # 核心执行脚本 (包含数据抓取逻辑)
└── references/
    └── personas/                 # 通用化人设提示词
        ├── buffett.md
        ├── wood.md
        └── druckenmiller.md

investment-committee/
├── SKILL.md                      # This file (architecture description)
├── TROUBLESHOOTING.md            # Troubleshooting
├── requirements.txt              # Dependencies (google-genai, yfinance)
├── scripts/
│   └── run_committee.py          # Core execution script (includes data fetching logic)
└── references/
    └── personas/                 # Generalized persona prompts
        ├── buffett.md
        ├── wood.md
        └── druckenmiller.md