AI Berkshire Skills: The Complete Guide to the I-6 Investment Research Framework

AI Berkshire ships with six core skills (the I-6 framework), not sixteen, covering investment research, industry analysis, financial data validation, earnings review, portfolio monitoring, and supply-chain bottleneck detection.

The xbtlin/ai-berkshire repository is an open-source investment research engine that codifies the workflows of four investment masters—Buffett, Munger, Duan Yong-ping, and Li Lu—into executable, auditable AI skills. While often mistakenly referred to as containing "16 skills," the platform actually implements the I-6 skill collection, six specialized Markdown-based skill packages that form a complete end-to-end research workflow.

The I-6 Skill Architecture: Six Core Capabilities

The "I-6" designation refers to six investment research skills, not sixteen. Each skill is defined as a plain-text Markdown file under the skills/ directory and automatically synchronized to a Codex-compatible package under codex-skills/ via scripts/sync-codex-skills.py. This dual structure ensures both human readability and compatibility with Claude Code and OpenAI Codex.

/investment-research

The flagship skill orchestrates full-cycle investment research using the "four masters" framework. Defined in skills/investment-research.md, this skill guides users through business quality assessment, management evaluation, and valuation rigor. It internally invokes /financial-data and /earnings-review for data gathering before producing a final investment thesis.

/industry-research

This skill performs industry-level analysis, market-size sizing, and "civilization-evolution" framing to identify long-term trends. The source definition lives at skills/industry-research.md and includes specific data source requirements for sector classification and competitive landscape mapping.

/financial-data

The data-validation backbone of the system, defined in skills/financial-data.md. This skill enforces the Financial Data Discipline: dual-source verification (US equities: Macrotrends + StockAnalysis; HK equities: AASTocks + Macrotrends; China A-shares: Eastmoney + 巨潮) and exact-arithmetic validation via tools/financial_rigor.py to eliminate LLM "mental math" errors.

/earnings-review

Handles deep earnings-report reading, data reconciliation, and post-earnings commentary generation. The skill definition at skills/earnings-review.md specifies transcript ingestion, KPI extraction, and cross-validation against financial data sources to ensure the <1% error tolerance required by the platform's rigor standards.

/portfolio-review

Provides portfolio-level monitoring, risk aggregation, and rebalancing recommendations. Located at skills/portfolio-review.md, this skill aggregates position-level data, calculates concentration risks, and flags deviations from target allocations using the standardized risk framework defined in AGENTS.md.

/bottleneck-hunter

A specialized skill for global supply-chain bottleneck detection and arbitrage opportunity scouting. Defined in skills/bottleneck-hunter.md, this scans for capacity constraints, freight cost anomalies, and inventory distortions across global markets to identify preemptive investment opportunities.

How AI Berkshire Skills Are Structured

Each skill follows a strict architectural pattern to ensure reproducibility and auditability:

Skill Definition – The canonical source is a Markdown file (skills/*.md) containing the workflow, required data sources, validation steps, and exact CLI commands.

Codex Generation – The scripts/sync-codex-skills.py utility reads the Markdown source and renders a Codex-compatible package (codex-skills/<skill>/SKILL.md), keeping Claude Code and Codex users synchronized with the canonical definitions.

Tooling Backbone – Core Python utilities under tools/ provide the computational rigor. tools/financial_rigor.py performs exact-arithmetic verification of market-cap and valuation calculations, while tools/report_audit.py generates embedded audit tables for every output.

Using the AI Berkshire Skills: CLI Examples

After installation via scripts/install-codex-skills.sh, invoke any of the six skills using the ai-berkshire CLI:

# 1. Full investment research on a US ticker

ai-berkshire /investment-research --ticker AAPL

# 2. Industry analysis for the semiconductor sector

ai-berkshire /industry-research --sector semiconductor

# 3. Pull and validate financial data for a Chinese A-share

ai-berkshire /financial-data --ticker 600519

# 4. Review latest earnings for a US stock

ai-berkshire /earnings-review --ticker MSFT

# 5. Generate portfolio health check

ai-berkshire /portfolio-review --holdings portfolio.csv

# 6. Scan for global supply-chain bottlenecks in AI hardware

ai-berkshire /bottleneck-hunter --focus AI

Each command produces a Markdown report in the working directory containing embedded audit tables generated by tools/report_audit.py, with traceability links back to the originating skill definition.

Data Discipline and Validation Rules

Every AI Berkshire skill adheres to the two-source rule mandated by the Financial Data Discipline:

• US Equities: Macrotrends + StockAnalysis
• HK Equities: AASTocks + Macrotrends
• China A-Shares: Eastmoney + 巨潮

All calculations use tools/financial_rigor.py for exact arithmetic, avoiding floating-point approximations that could compound into material valuation errors. This discipline ensures that outputs meet the <1% error tolerance required for investment-grade research.

Summary

• AI Berkshire implements six skills (I-6), not sixteen: /investment-research, /industry-research, /financial-data, /earnings-review, /portfolio-review, and /bottleneck-hunter.
• Each skill is defined as a Markdown file under skills/ and synchronized to codex-skills/ via scripts/sync-codex-skills.py.
• The platform enforces dual-source verification and exact-arithmetic calculation through tools/financial_rigor.py.
• Skills are composable: /investment-research chains /financial-data and /earnings-review for comprehensive analysis.
• All outputs include automated audit trails generated by tools/report_audit.py.

Frequently Asked Questions

Does AI Berkshire have 16 skills or 6 skills?

AI Berkshire contains six skills, not sixteen. The confusion often stems from the "I-6" designation (Investment-6), which refers to the six core skill packages. These are /investment-research, /industry-research, /financial-data, /earnings-review, /portfolio-review, and /bottleneck-hunter, as defined in the skills/ directory.

How do I install and use AI Berkshire skills?

Install the skill suite by running scripts/install-codex-skills.sh from the repository root. Once installed, invoke any skill via the ai-berkshire CLI followed by the slash-command and relevant parameters (e.g., --ticker or --holdings). Each execution produces a Markdown report with embedded audit tables.

What is the difference between the skills/ and codex-skills/ directories?

The skills/ directory contains the canonical Markdown definitions readable by humans and version-controlled as the source of truth. The codex-skills/ directory contains generated Codex-compatible packages produced by scripts/sync-codex-skills.py. The sync script ensures that Codex and Claude Code users receive the same logic as the canonical Markdown files.

How does AI Berkshire ensure financial data accuracy?

The platform enforces the Financial Data Discipline: requiring dual-source verification for all data points (e.g., Macrotrends + StockAnalysis for US equities) and using tools/financial_rigor.py for exact-arithmetic calculations. This prevents LLM hallucination in quantitative analysis and maintains a <1% error tolerance threshold for all generated reports.

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