AI Berkshire is a collection of value-investing research skills that run in Claude Code (as commands) and Codex (as skills) from one shared workflow. It encodes the methods of four investors (Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu) into structured research procedures, and its content is written in Chinese with English and Japanese translations.
The skills force a decision rather than a balanced essay: each analysis ends in pass, fail, or a gray zone, with price ranges and tiered position suggestions. The four investor perspectives are set up to challenge each other (business model, financial valuation, inversion and risk, and long-term certainty), each producing its own score. Built-in checks include an information-richness grade (A, B, or C), a Munger-style inversion test of failure scenarios, an eight-item fast veto list, a contrarian check, and a rule to mark a gray zone instead of guessing when data is thin.
Team skills such as investment-team and earnings-team have a Team Lead dispatch four investor-perspective agents in parallel, each searching the web, cross-checking data, and reaching its own conclusion before the lead combines them. Lighter skills skip the agent layer and call tools directly. A Python financial-rigor tool recalculates market capitalization, valuation ratios, three-scenario target prices, and Benford's-law anomalies with exact decimal arithmetic, and cross-validates figures from at least two independent sources.
The skills follow the same checklist structure for every company, so results stay comparable across companies and across reruns. The repository also carries a large archive of generated research reports grouped by company and theme.
Features
- Twenty skills: grouped into deep research, earnings analysis, industry screening, portfolio management, and thinking tools
- Four-investor framework: Buffett, Munger, Duan Yongping, and Li Lu perspectives scored separately and set against each other
- Forced verdicts: pass, fail, or gray zone with price ranges and aggressive, balanced, or conservative position suggestions
- Multi-agent teams: a Team Lead runs four investor agents in parallel on one company or one earnings report and synthesises their findings
- Bias controls: information-richness grading, inversion testing, an eight-item veto list, a contrarian check, and explicit gray-zone marking
- Financial rigor tool: exact-decimal market-cap and valuation checks, multi-source cross-validation, three-scenario pricing, and Benford's-law detection
- Industry screening: value-chain scans, a funnel from the whole market down to three finalists, a seven-metric quality screen, and supply-chain bottleneck hunting
- Portfolio discipline: portfolio review, investment-thesis tracking, thesis-drift comparison, income-stock analysis, and fast attribution of sharp price moves
- Private-company research: a dedicated skill for unlisted companies with sparse public information
- Article writing: author, editor, and reader agents collaborate to turn research into publishable WeChat articles
