Codex 搭配下面这些 Skill,可以帮你搭建一套专业的美股投研系统: 1、ai-berkshire:这个skill结合巴菲特、芒格、段永平等投资大佬的经典方法论,让codex具有一套宏观的研究框架 :https://github.com/xbtlin/ai-berkshire… 2、UZI-Skill:让codex具备深度个股研究能力,从基本面、技术面、估值多角度分析特定美股 :https://github.com/wbh604/UZI-Skill… 3、QuantDinger:将投研结论落地为可测试、可跟踪的交易方案,用数据验证你的交易想法 :https://github.com/brokermr810/QuantDinger…

When paired with the following Skills, Codex can help you build a professional US stock investment research system:

1. ai-berkshire: Combines the classic methodologies of investment legends like Buffett, Munger, and Duan Yongping, giving Codex a macro-level research framework: https://github.com/xbtlin/ai-berkshire
2. UZI-Skill: Enables Codex to conduct deep individual stock research, analyzing specific US stocks from fundamental, technical, and valuation perspectives: https://github.com/wbh604/UZI-Skill
3. QuantDinger: Turns research conclusions into testable, trackable trading plans, using data to validate your trading ideas: https://github.com/brokermr810/QuantDinger

xbtlin/ai-berkshire

Source: https://github.com/xbtlin/ai-berkshire

中文 | English

GitHub Trending (https://trendshift.io/repositories/63696)

AI Berkshire - A Value Investment Research Framework for the AI Era

“Price is what you pay, value is what you get.” — Warren Buffett

Redefining the depth and efficiency of investment research with AI.

AI Berkshire is a collection of Skill packages compatible with both Claude Code and Codex. It systematizes and structures the methodologies of four value investing masters — Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu — enabling professional-grade investment research through AI Agents.

One person + Claude Code / Codex = an entire research team.

Real Track Record · Why Not Just Ask AI? · Skills Overview · Quick Start · Research Reports · Design Philosophy

Real Track Record

Not just theoretical. This framework is backed by a real-money investment system.

2024 Full Year Return: +69.29%

2025 Full Year Return: +66.38%

Comparison with Major Indices

2024 Excess Return: Beat S&P 500 by 46 percentage points, beat Hang Seng Index by 52 percentage points. 2025 Excess Return: Beat S&P 500 by 50 percentage points, beat Hang Seng Index by 39 percentage points.

Cumulative two-year live trading profit exceeded 1.46 million CNY, significantly outperforming major global indices for two consecutive years.

Disclaimer: Past performance does not guarantee future results. Screenshots are from a real Futu Securities account.

Why Not Just Ask AI?

Of course, you can directly ask Claude: “Help me analyze whether Pinduoduo is worth buying.” You’ll get a balanced analysis filled with “on one hand… on the other hand…”, ending with “Investing involves risk, please make your own judgment.”

This kind of analysis looks correct but is useless for making decisions.

AI Berkshire doesn’t solve the “can you analyze” problem; it solves the quality of analysis and decision-making discipline problem. Here are the core differences:

1. Forced Conclusion, No Fence-Sitting

If you ask AI directly, you get a “balanced” analysis that pleases both sides. AI Berkshire forces an output: Pass / Fail / Grey Zone, with specific price ranges and tiered suggestions.

Normal AI response: “Pinduoduo has growth potential but also faces competitive pressures. Investors need to weigh…”

AI Berkshire output:

Strategy	Suggestion	Price Range
Aggressive	Can build 20% position at current price	$95-105
Moderate	Build position after buyback policy clear	$85-95
Conservative	Doesn’t meet 10-year certainty criteria, wait	—

Mirror Test: If you can’t explain it in 5 sentences = don’t buy. No exceptions.

2. Four Masters’ Perspectives in Tension, Not Single Analysis

It’s not simply “analyze with Buffett’s method.” Four perspectives generate real contradictions and tensions — using Pinduoduo as an example:

• Duan Yongping (Business Model): Good business, C2M model hard to replicate → Score 3.7/5
• Warren Buffett (Financial Valuation): PE ex-cash only 6.3x, cash printer → Score 4.4/5
• Charlie Munger (Inversion Thinking): Moat is shallower than imagined; Douyin did 4 trillion GMV in 3 years → Score 3.5/5
• Li Lu (Long-term Certainty): Management culture has hidden risks, uncertain 10 years from now → Score 2.0/5

Buffett says “really cheap,” Li Lu says “if unsure, don’t buy” — this conflict is the real state of investment decisions. A single prompt can’t create this multi-perspective tension, which is precisely the key to avoiding blind spots.

3. Structured Anti-Bias Mechanisms

The most dangerous thing about AI isn’t giving wrong answers, but giving an answer that looks right but falls apart under scrutiny. AI Berkshire embeds multiple layers of “fraud prevention” in its process:

4. Financial Data Accuracy

LLM mental math is unreliable. Getting PE wrong by a decimal point or confusing HKD and RMB for market cap can lead to wrong investment decisions.

Real Case: When analyzing Tencent, market cap data from different sources had “HKD billions” and “RMB billions” units. AI Berkshire handles it:

# Market cap manual check: stock price × total shares, compare with reported data
python3 tools/financial_rigor.py verify-market-cap \
    --price 510 --shares 9.11e9 --reported 4.65e12 --currency HKD
# ✅ Pass, deviation only 0.08%

All calculations use Python decimal.Decimal (exact decimal), not float. Key data is cross-verified from at least 2 independent sources.

5. Reproducible Research Process

If you ask AI directly, each output has different format, depth, and coverage — today’s Tencent analysis has a moat score, tomorrow’s Meituan analysis might forget it.

AI Berkshire ensures: Same input → consistent structure, consistent depth of output. This means you can:

• Compare 7 companies horizontally with identical scoring standards
• Re-analyze the same company six months later and directly compare changes
• Align research results among team members

Real output — 7 companies screened with the same standard Checklist:

Company	Pass?	Circle of Competence	Good Business	Moat	Management	Margin of Safety	Overall
Moutai	✅ Pass	★★★★★	★★★★★	★★★★★	★★★☆☆	★★★★☆	4.7
Tencent	✅ Pass	★★★★☆	★★★★★	★★★★★	★★★★★	★★★★☆	4.7
NVIDIA	✅ Conditional	★★★★☆	★★★★★	★★★★★	★★★★★	★★★☆☆	4.3
Meituan	✅ Conditional	★★★★☆	★★★★☆	★★★★☆	★★★★☆	★★★★☆	4.0
Kuaishou	✅ Conditional	★★★☆☆	★★★★☆	★★★★☆	★★★★☆	★★★★★	4.0
Pinduoduo	❓ Grey Zone	★★★★☆	★★★★☆	★★★☆☆	★★★☆☆	★★★★★	3.8
Pop Mart	❓ Grey Zone	★★★☆☆	★★★★☆	★★★★☆	★★★★★	★★★☆☆	3.7

6. Multi-Agent Parallel = Multiplied Research Depth

/investment-team launches 4 independent Agents to study a single company simultaneously. Each Agent independently searches the web, cross-validates data, and gives its own conclusion. This isn’t splitting one prompt into four parts — it’s 4 “analysts” each doing complete research, with the Team Lead synthesizing.

One person asking AI directly has one context window. 4 Agents running in parallel equals 4x the search volume, 4x the information sources, and 4 independent perspectives.

┌───────────────────────────────────────────────────────────────┐
│                    Team Lead (You)                              │
│              Coordination · Summary & Judgment                 │
├──────────┬──────────┬──────────────┬─────────────────────┤
│ Agent 1   │ Agent 2   │ Agent 3       │ Agent 4                │
│ Business  │ Financial │ Industry      │ Risk & Management       │
│ Model     │ Valuation │ Competition   │                        │
│ Duan Yongping│ Buffett │ Munger Perspective│ Li Lu Perspective   │
│ Perspective│ Perspective│               │                        │
└──────────┴──────────┴──────────────┴─────────────────────┘
          ↓             ↓               ↓                  ↓
              Parallel Research, Real-Time Progress Updates
                            ↓
                     Final Comprehensive Report

In One Sentence

When ordinary people ask AI, they get “analysis that looks correct”; with AI Berkshire, you get “an investment research report you can actually use to make decisions.”

Overall Architecture

Source: assets/architecture.mmd (Mermaid editable source)

Three-Layer Design Philosophy:

• Skill Layer: Abstracts “what you want to do” into 18 clear entry points — deep research, earnings analysis, industry screening, portfolio management, thinking tools — choose by scenario.
• Agent Layer: Inside each skill, 4 Agents work in parallel — they search independently, judge independently, challenge each other, and are finally synthesized by the Team Lead.
• Tool Layer: Precise calculations, real-time retrieval, report spot-checks — ensuring every report’s data rigor is verifiable.

Skills Overview (18 Skills)

🔬 Deep Research

📊 Earnings Analysis

🏭 Industry Screening

📈 Portfolio Management

🧠 Thinking Tools

Quick Start

1. Install AI Client

This repository maintains the same canonical workflow and provides both Claude Code commands and Codex skills. Install according to the client you use.

Claude Code users:

npm install -g @anthropic-ai/claude-code

Codex users:

# macOS / Linux
curl -fsSL https://chatgpt.com/codex/install.sh | sh

# or using npm
npm install -g @openai/codex

# or using Homebrew
brew install --cask codex

# verify installation
codex --version

Windows users can use the official PowerShell installation command: powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex".

If codex --version outputs a version number, you can proceed to install this project’s Codex skills.

Reduce Authorization Prompts

These skills frequently invoke tools. By default, Claude Code asks for authorization confirmation each time. This behavior comes from Claude Code’s client permission mechanism, which cannot be changed by this repository.

If you trust the current workflow and are running in a trusted environment, you can start Claude Code with the skip-permission-confirmation mode:

claude --dangerously-skip-permissions

Note: This mode disables Claude Code’s tool approval guard. Only use it if you trust the repository, commands, and working directory.

2. Install Skills

Claude Code users:

# Clone repo
git clone https://github.com/xbtlin/ai-berkshire.git

# Copy skills to Claude Code's global commands directory
cd ai-berkshire
./scripts/install-claude-commands.sh

Codex users:

# Clone repo
git clone https://github.com/xbtlin/ai-berkshire.git

# Generate and install Codex skills to ~/.codex/skills
cd ai-berkshire
./scripts/install-codex-skills.sh

# Optional: Install Codex slash prompts to ~/.codex/prompts
# to get a similar /investment-research experience to Claude Code
./scripts/install-codex-prompts.sh

The repository maintains three entry points simultaneously: skills/*.md are the Claude Code command source files; codex-skills/*/SKILL.md are Codex skill packages generated from skills/*.md by scripts/sync-codex-skills.py; codex-prompts/*.md are an optional Codex slash prompt compatibility layer.

3. Usage

In Claude Code:

# Deep Research
/investment-research Tencent
/investment-team Meituan
/management-deep-dive Wang Xing Meituan
/private-company-research SpaceX
/deep-company-series Pinduoduo

# Earnings Analysis
/earnings-review Tencent 2025Q4
/earnings-team PDD 2025 Annual Report

# Industry Screening
/industry-research Nuclear Power
/industry-funnel AI Computing Power
/quality-screen Hang Seng Index Constituents
/bottleneck-hunter AI Infrastructure
/investment-checklist Moutai, NVIDIA, Apple

# Portfolio Management
/portfolio-review Tencent 30%, Meituan 20%, Moutai 20%, Cash 30%
/thesis-tracker Pinduoduo
/news-pulse Tencent

# Thinking Tools
/dyp-ask Where is Pinduoduo's moat really?
/wechat-article Meituan

In Codex: After installation, restart Codex and describe the task using the skill name, for example:

Use investment-research to research Tencent
Use earnings-review to analyze PDD 2025 Annual Report
Use industry-funnel to screen AI Computing Power
Use bottleneck-hunter to scan AI infrastructure bottlenecks
Use wechat-article to write a Meituan investment article

If you installed Codex slash prompts, restart Codex and you can also search for these prompts in the / menu. Codex’s official custom prompt entry is usually displayed as prompts:, for example:

/prompts:investment-research Tencent

Detailed Skill Descriptions

1. /investment-research — Four Masters Comprehensive Analysis

The most comprehensive single-company deep research framework. Executed in seven sequential modules:

Data Collection → Business Nature (Duan Yongping) → Moat (Buffett) → Inversion Thinking (Munger) → Management Assessment (Duan + Buffett) → Civilization Trends (Li Lu) → Valuation & Margin of Safety

Key Features:

• AI Research Bias Awareness Mechanism (A/B/C Information Richness Rating)
• Multi-source cross-validation of key data (market cap manual calculation check, at least 2 independent sources)
• Four Masters’ “probing questions” throughout
• Three-scenario valuation (Bull / Base / Bear) + Reverse DCF

Sample Output Excerpt:

Comprehensive Decision Memo

Dimension	Conclusion	Confidence
Business Quality (Duan Yongping)	Excellent: Platform business, bilateral network effects, marginal cost approaches zero	★★★★★
Moat (Buffett)	Wide and widening: Network effects + Switching costs + Economies of scale triple stack	★★★★☆
Management (Duan + Buffett)	Excellent: Founder-led, strong capital allocation discipline	★★★★☆
Biggest Risk (Munger)	Regulatory policy uncertainty, new business losses dragging overall profit	★★★☆☆
Civilization Trend (Li Lu)	Aligns with digital consumption trend, but not a “civilization-level paradigm shift”	★★★★☆
Valuation (Buffett + Duan)	Current PE 18x, at historical median-low, some margin of safety	★★★★☆

Duan Yongping: “The essence of this business is connecting consumers and merchants, making money from efficiency gains. The hallmark of a good business: more users bring more merchants; more merchants bring more users. Once the flywheel starts spinning, it’s hard to stop.”

Charlie Munger: “Think upside down — if this company disappeared tomorrow, what would users and merchants do? If the answer is ‘they’d quickly find a substitute,’ then the moat isn’t deep enough. If the answer is ‘life would become very inconvenient,’ that’s worth paying attention to.”

2. /investment-team — Multi-Agent Research Team

Launches 4 AI Agents to research in parallel, simulating a real research team collaboration. Each Agent searches independently, analyzes independently, and gives its own score. The Team Lead then synthesizes and judges.

Sample Output Excerpt:

One-Sentence Conclusion

Meituan is the absolute leader in China’s local services market, with a multi-layered network effect moat. Current valuation is at historically low levels, offering significant long-term investment value. Recommended to build positions on dips.

Four-Dimension Score Summary

Dimension	Framework	Score	Core Judgment
Business Model & Moat	Duan Yongping	★★★★☆	Strong bilateral network effects; food delivery + in-store form a flywheel
Financial & Valuation	Buffett	★★★★☆	Core business margins consistently improving; valuation at historical lows
Industry & Competition	Munger	★★★☆☆	Douyin encroaching on in-store business; competitive landscape may worsen
Risk & Management	Li Lu	★★★★☆	Wang Xing has excellent strategic vision, but new business cash burn needs vigilance

Overall Score: 3.8 / 5

Investment Recommendation

Strategy	Suggestion	Price Range (HKD)
Aggressive	Build 30% position at current price	120-140
Moderate	Build position after pullback to 100-110	100-120
Conservative	Wait for quarterly results to confirm margin trend before entry	<100

3. /investment-checklist — Buffett’s Pre-Purchase Checklist

Six-stage quick screen to decide in 10 minutes whether a company is worth deeper research:

Stage 1: Circle of Competence (Do I understand it?)
    ↓ Pass
Stage 2: Good Business (What are its economic characteristics?)
    ↓ Pass
Stage 3: Moat (Is the competitive advantage deep?)
    ↓ Pass
Stage 4: Management (Are they trustworthy?)
    ↓ Pass
Stage 5: Margin of Safety (Is the price attractive?)
    ↓ Pass
Stage 6: Decision Discipline (Is it rationality or FOMO?)
    ↓ Pass
✅ Mirror Test

Supports multi-company comparison — screen multiple targets at once:

/investment-checklist Tencent, Alibaba, Meituan, Pinduoduo

Sample Output Excerpt:

Mirror Test

“I am buying Tencent at 380 HKD because:

1. The essence of this business is social network + digital content platform, I understand it.
2. Its moat is the social relationship network of 1.2 billion users, and it’s widening.
3. Management is Pony Ma — low-key, pragmatic, excellent capital allocation — trustworthy.
4. Current price represents a 20% discount to intrinsic value — some margin of safety.
5. Even if I’m wrong, downside risk is controllable because net cash on books exceeds 200 billion RMB, game cash flow is strong.“

✅ Passes Mirror Test

Can’t explain it in 5 sentences = don’t buy. No exceptions.

4. /industry-research — Full Industry Chain Panoramic Scan

Starting from an investment theme, complete a full industry chain panoramic study:

Build Investment Logic Chain → Industry Panoramic Map → Global Listed Company Scan → Four Masters Analysis of Leading Companies in Each Segment → Portfolio Allocation Recommendation

Sample Output Excerpt:

Investment Logic Chain: Nuclear Power

Underlying trend: AI data center electricity demand explosion + Carbon neutrality goals → Leads to: Surge in demand for stable, clean baseload power → Creates: Certain demand for nuclear restart/new build/SMR → Benefits: Uranium mining → Fuel processing → Equipment manufacturing → Operators

Recommended Portfolio

Tier	Allocation	Targets	Segment	Core Logic
Core	50%	CGN Power, Cameco	Operations + Uranium	Highest certainty
Satellite	30%	China National Nuclear Power, Dongfang Electric	Operations + Equipment	Beneficiary of domestic substitution
Options	15%	NuScale, Nano Nuclear	SMR	High risk, high reward
ETF	Alternative	URA, URNM	Full chain	Lazy option

5. /industry-funnel — Industry Funnel Screening

Starting from an industry/direction, Full Market → ≤10 Companies → 3 Companies layer-by-layer selection:

Full Market Scan (active + gainers + top 30 by market cap union, 30-60 companies)
    ↓
Rough Screen using 5 value investing hard criteria → ≤10 companies
    ↓
Detailed Analysis (300-500 words each) on ≤10 companies
    ↓
Final Selection (based on portfolio complementarity, not top 3 by score)
    Four Masters Deep Analysis on 3 companies (800-1200 words each)
    ↓
Recommended Portfolio (Core / Satellite / Options) + Action Signals

Key Features:

• Each layer has clear keep/discard criteria; eliminated targets leave a reason for elimination (not a black box)
• Final 3 selected by “portfolio complementarity” (high certainty + medium upside + high upside), not top 3 by score
• Mandatory “future IPO candidates” list to avoid missing key private market players
• AI bias awareness mechanism: addresses leader bias / English bias / narrative bias / listing bias

Difference from /industry-research:

• industry-research focuses on industry chain structure and panorama (segmented by link)
• industry-funnel focuses on stock screening funnel (layer-by-layer selection from full market to 3 companies)

Test Run: 4 AI Sub-Tracks in Parallel (2026-05-09):

Key Finding: The biggest winners in the AI application layer aren’t AI-native companies, but established giants with channels + data + workflow embeddedness — echoing the 1995-2000 internet bubble “selling shovels” pattern (Amazon and Apple won, Pets.com lost).

Full reports: AI Computing Power · AI Models · AI Applications · AI Infrastructure Power

6. /private-company-research — Deep Research on Unlisted Companies

A “detective-style” research framework designed for unlisted companies with scarce information:

Key Differentiators:

• Financial Data Piecing Together: Piecing together data from prospectuses, parent company filings, funding news, and industry data
• Confidence Labeling: Each data point labeled 🟢 High / 🟡 Medium / 🔴 Low confidence
• Multiple Valuation Methods Cross-Check: Funding valuation method + Comparable company analysis + DCF + Terminal value back-calculation
• Exit Path Analysis: Full evaluation of IPO / M&A / secondary transfer paths

Sample Output Excerpt:

Company Snapshot: SpaceX

Item	Content
Latest Valuation	~$350B (secondary market 2025) 🟡
Estimated Revenue	~$13B (2024) 🟡
Starlink Users	4M+ (end of 2024) 🟢
Launches	100+ per year (2024) 🟢

Valuation Judgment

Method	Valuation Range	Notes
Latest Funding Round	$350B	Secondary market quote, with liquidity premium
Comparable Company Method	$200-280B	Benchmarking telco + aerospace + defense
DCF (Base Case)	$250-350B	Assumes Starlink $30B revenue by 2027
Terminal Value Back-Calculation	$400-600B	Assumes Starlink becomes global telecom infrastructure

Comprehensive Reasonable Valuation Range: $250B - $400B

7. /news-pulse — Stock Price Move News Attribution

A rapid intelligence response Skill designed for “when the stock surges/crashes, quickly understand what happened.” Not deep research; it’s a 10-15 minute rapid attribution — to avoid getting caught up in rumors or panic stop-loss during holdings volatility.

Key Differentiators:

• 4-Dimensional Parallel Reconnaissance: Company events / Regulatory policy / Industry competitors / Market sentiment (sell-side + influencers + southbound capital)
• Attribution Prioritized Over Listing: Not just listing all news, but determining “which event matches this price move”
• Mandatory Nature Judgment: Value event / Sentiment fluctuation / True cause unknown / Mixed — “True cause unknown” is the most valuable output (possible insider front-running)
• Clear Action Suggestions: Whether to trigger deep research, whether to review thesis, or just observe, etc.

Difference from Other Skills:

Sample Output Excerpt (Tencent live test 4/17-5/01, -10.47% in 14 days):

One-Sentence Attribution

This -10.47% decline is approximately 70-80% driven by capital flow + sentiment factors (buyback blackout period + southbound capital reduction + sector beta + AI narrative taken over) and 20-30% by digesting the doubled AI investment. No fundamental negative catalysts. Sell-side maintains consensus Strong Buy. Nature assessment: Liquidity + sentiment-driven correction, not a value event.

Attribution Table

Candidate Explanation	Estimated Contribution	Confidence
Buyback blackout period disappearing (structural, before 5/13 earnings)	-3% ~ -4%	High
Southbound capital turning to net sell Tencent	-2% ~ -3%	High
AI narrative taken by competitors (DeepSeek V4/Qwen3.6/Yue’an 1T)	-1% ~ -2%	Medium
Sector/macro beta (oil + geopolitics + Fed Warsh hawkish)	-2% ~ -3%	High
Pre-Q1 earnings risk-off	-1% ~ -2%	Medium
Fundamental deterioration	0%	Very High (excluded)

Nature Judgment: ✅ Mixed

70% capital flow/sentiment + 20% AI long-term narrative concern + 10% pre-earnings uncertainty

Key Counter-Evidence: Duan Yongping sold Tencent puts on 4/8 (bullish); 24 sell-side consensus Strong Buy; NetE