mcpField
I
investor-panel

investor-panel

github.com/wbh604Data & databases7.4k181 installsUpdated 2w ago
Repository

About investor-panel

investor-panel is a skill for stock-analysis workflows that generates 50 investor Signals from stock data and investing methodologies, then aggregates their votes.

Key features

  1. 01Reads scoring dimensions and raw data
  2. 02Generates Signals using investor methodologies
  3. 03Pre-filters traders by style range
  4. 04Validates strict JSON output
  5. 05Aggregates consensus and vote distributions

Use cases

  1. 01Creating an investor panel for a ticker
  2. 02Filtering stocks outside a trader's style
  3. 03Summarizing bullish, neutral, and bearish signals

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/wbh604/uzi-skill --skill investor-panel

After installing, type/investor-panelin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

assets/
investor-cards.json
references/
group-a-classic-value.md
group-b-growth.md
group-c-macro-hedge.md
group-d-technical.md
group-e-china-value.md
group-f-china-youzi.md
group-g-quant.md
group-i-serenity.md
quotes-knowledge-base.md
serenity-voice.md

Details

From the project README

Investor Panel · 50 贤评审团

调用上下文

读取以下输入:

  • .cache/{ticker}/dimensions.json — 19 维评分
  • .cache/{ticker}/raw_data.json — 原始数据
  • scripts/lib/investor_db.py — 65 人元数据
  • scripts/lib/seat_db.py — 22 位游资射程规则

输出:

  • .cache/{ticker}/panel.json — 50 个 Signal + 投票统计

严格输出格式(Pydantic Signal,抄自 ai-hedge-fund)

每个投资者必须返回严格 JSON:

{
  "investor_id": "buffett",
  "name": "巴菲特",
  "group": "A",
  "avatar": "avatars/buffett.svg",
  "signal": "bullish | neutral | bearish",
  "confidence": 87,
  "score": 82,
  "verdict": "强烈买入 | 买入 | 关注 | 观望 | 等待 | 回避 | 不达标 | 不适合",
  "reasoning": "1-3 句具体逻辑",
  "comment": "用该投资者语言风格的金句 1-2 句",
  "pass": ["..."],
  "fail": ["..."],
  "ideal_price": 16.20,
  "period": "3-5 年"
}

Confidence 校准规则:

  • 85-100:核心方法论硬指标全部命中或全部不命中
  • 60-84:多数命中
  • 30-59:部分命中、需要等待信号
  • 0-29:方法论不适用此股 / 信息不足

执行步骤

Step 1: 加载元数据
from lib.investor_db import INVESTORS, by_group
from lib.seat_db import SEATS, is_in_range
Step 2: 对每位投资者
  1. 取出 fields 白名单
  2. 从 dimensions.json 提取相关字段
  3. 读取该投资者所在 group 的 reference 文件(按需)
  4. 用该投资者的方法论 + 语言样本生成 Signal(Claude 自己生成)
  5. 校验 JSON 合法性
Step 3: 游资射程预过滤(F 组特殊)

对 22 位游资,先用 is_in_range(nickname, ticker_features) 判断是否在射程内:

  • 在射程 → 正常评分
  • 不在射程 → signal: "neutral", verdict: "不适合", confidence: 90, comment: "{nick}的射程是{style},这只票不在风格内。"
Step 4: 汇总投票
{
  "panel_consensus": (bullish_count / 50) * 100,
  "vote_distribution": Counter(verdict for i in investors),
  "signal_distribution": Counter(signal for i in investors),
  "investors": [...]
}

7 大流派详细方法论

按需读取下列 references:

组文件人数
A 经典价值references/group-a-classic-value.md6
B 成长投资references/group-b-growth.md4
C 宏观对冲references/group-c-macro-hedge.md5
D 技术趋势references/group-d-technical.md4
E 中国价投references/group-e-china-value.md6
F 游资references/group-f-china-youzi.md22
G 量化系统references/group-g-quant.md3

📚 语料库 (必读)

每次生成 comment 之前必须读 references/quotes-knowledge-base.md 查找该投资者的真实公开原话和"风格"字段。这是知识库 single source of truth。

语言风格守则

每位投资者的 comment 字段必须像他本人:

  • 巴菲特:温和、引用奥马哈、用"我们"
  • 芒格:刻薄、反向思维、引用心理学偏误
  • 索罗斯:哲学化、提"反身性"
  • 章盟主:豪迈、提"格局"、不谈细节
  • 赵老哥:直接、谈"题材"、谈"二板"
  • 段永平:朴素、问"商业模式""人""价格"
  • 陈小群:江湖气、谈"分歧""一线天""核按钮"

每组 reference 文件末尾有 3-5 句真实公开语录作为 few-shot。

完成检查

  • panel.json 包含 50 个 Signal
  • 每个 Signal 字段齐全
  • 22 位游资里至少有 N 位返回"不适合"(除非这只票是热门题材龙头)
  • panel_consensus / vote_distribution / signal_distribution 三个汇总字段已计算

Comments

Markdown supported