How to Use the /investment-research Skill in AI Berkshire: Complete Guide

TLDR: Invoke /investment-research followed by a company ticker or name (e.g., /investment-research 腾讯) to trigger a seven-module parallel analysis pipeline that synthesizes investment perspectives from Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu, outputting a structured markdown report with confidence ratings and financial verification.

The /investment-research skill in the xbtlin/ai-berkshire repository implements the flagship "四大师综合分析" (Four Masters Comprehensive Analysis) workflow. This open-source tool orchestrates a multi-agent research pipeline defined in skills/investment-research.md that combines first-principles business analysis with rigorous financial verification to generate institutional-grade investment memos.

Installation and Setup

Before invoking the skill, ensure it is registered in your AI coding environment. The repository provides automated installers for both Claude Code and OpenAI Codex.

For Claude Code, run the bash installer to copy the skill into the global command directory:

bash scripts/install-claude-commands.sh

For Codex, the skill wrapper is auto-generated at codex-skills/investment-research/SKILL.md, linking the markdown source to the Codex runtime. Verify the skill appears in your command palette by checking the skill registry located in the installation target directory.

Basic Usage Commands

The skill accepts a company identifier (ticker or Chinese name) as its primary argument.

Command-Line Invocation

Execute the full deep-research workflow directly in your AI assistant chat:

# Basic analysis on Tencent

/investment-research 腾讯

# With optional scenario flag (requires custom argument parsing)

/investment-research 腾讯 --scenario three

Programmatic Access

Trigger the skill from Python scripts using the Codex runtime interface:

python3 -c "import codex; codex.run('investment-research', '腾讯')"

The Seven-Module Research Pipeline

The skill file at skills/investment-research.md (lines 13-92) defines a sequential yet parallelizable workflow. Each module corresponds to a specific investment master's philosophy.

Pre-Research Bias Check (Lines 13-31)

Before data collection, the skill rates the information richness of the target (A/B/C classification) and executes the "AI research bias self-awareness" checklist. This prevents over-confidence when data are scarce, forcing the system to annotate low-confidence sections explicitly.

Data Collection and Cross-Validation (Lines 35-53)

A background Task agent gathers raw financial data, market share statistics, and management biographies. All quantitative inputs are cross-validated by the financial-rigor tool at tools/financial_rigor.py (lines 55-92) before entering the analysis phase.

Business Essence Analysis (Duan Yongping)

Based on lines 20-25 of the skill definition, this module applies first-principles questioning regarding customer value, repeatability of purchase, competitive dynamics, and management decision-making quality.

Moat Assessment (Buffett)

Identifies durable competitive advantages using Buffett's economic fortress framework, evaluating pricing power, switching costs, and network effects.

Reverse Thinking (Munger)

Lines 46-48 force "what-if-the-company-dies?" scenario analysis, inverting the investment thesis to identify potential failure modes before confirming bullish assumptions.

Civilization Trend Evaluation (Li Lu)

Evaluates long-term macro forces and secular trends that could impact the business over decade-long horizons, filtering out companies swimming against historical currents.

Valuation and Safety Margin (Lines 81-86)

Computes multi-scenario target prices using exact arithmetic. The rigor tool performs calculations with Python's decimal.Decimal to eliminate floating-point errors in market-cap and PE computations.

Parallel Agent Architecture

Internally, the skill launches four independent agents (one per master), each executing the full data-gather-analyze-report cycle simultaneously. A Team-Lead agent collates these parallel outputs, ensuring redundancy and cross-perspective validation (see the architecture diagram in README.md, lines 31-45).

This architecture means the Buffett agent evaluates moats while the Munger agent stress-tests failure modes concurrently, reducing total research time without sacrificing depth.

Financial Verification and Rigor

All numeric calculations undergo verification through tools/financial_rigor.py. For manual verification of specific figures, invoke the tool directly:

python3 ~/ai-berkshire/tools/financial_rigor.py verify-market-cap \
  --price 510 --shares 9.11e9 --reported 4.65e12 --currency HKD

The tool validates market-cap calculations, cross-references multi-source data, and embeds output logs as an appendix in the final report, guaranteeing reproducibility.

Output Structure and Interpretation

The skill emits a structured markdown report containing three critical components:

1. Information-richness rating (A/B/C) and confidence annotation at the document header
2. 综合决策备忘录 (Integrated Decision Memo) table summarizing each master's view (lines 50-60)
3. 镜子测试 (Mirror Test) verdict requiring a five-sentence justification paragraph that forces the AI to explicitly state whether the investment passes or fails based on the synthesized analysis

Summary

• The /investment-research skill combines four legendary investment frameworks into a single command.
• Install via scripts/install-claude-commands.sh or use the Codex wrapper at codex-skills/investment-research/SKILL.md.
• The workflow runs seven modules: bias check, data collection, business essence, moat analysis, reverse thinking, civilization trends, and valuation.
• Four parallel agents (one per master) execute simultaneously, coordinated by a Team-Lead agent.
• All financial calculations use decimal.Decimal for exact arithmetic and are validated by tools/financial_rigor.py.
• Output includes confidence ratings, a decision memo table, and a mandatory mirror-test verdict.

Frequently Asked Questions

How do I install the /investment-research skill in my Claude Code environment?

Run the installation script scripts/install-claude-commands.sh from the repository root. This copies the skill definition from skills/investment-research.md into Claude's global command directory, making /investment-research available as a native slash command. Verify installation by checking that the skill appears in your Claude Code command palette.

What data sources does the investment-research skill use?

The skill delegates data gathering to a background Task agent that aggregates financial statements, market share data, and management information. All quantitative inputs are cross-validated by the tools/financial_rigor.py utility before analysis. The system prioritizes primary sources and flags external data with low verification confidence in the final report.

How does the skill handle low-information scenarios?

During the pre-research phase (lines 13-31), the skill rates information richness as A, B, or C. If data is insufficient (C rating), the system activates the "AI research bias self-awareness" checklist, forcing explicit confidence annotations and preventing false precision. The final output will prominently display uncertainty warnings and avoid definitive buy/sell recommendations when data quality is poor.

Can I customize the analysis modules or add additional investment masters?

The skill architecture supports extension through the markdown definition in skills/investment-research.md. You can modify the seven-module sequence by editing the prompt templates between lines 20-92. To add a fifth master, create a new agent instance in the parallel architecture (documented in README.md, lines 31-45) and add corresponding lens-specific prompts to the skill file.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}