ai-berkshire

AI Berkshire is an investment research Skill set for Claude Code and Codex that breaks down the methodologies of four value investing masters—Buffett, Munger, Duan Yongping, and Li Lu—into structured workflows executed by AI Agents. The author's positioning is "one person + AI client = an investment research team." Core Approach The README emphasizes that what it solves is not "whether analysis can be done," but analysis quality and decision discipline: it forces output of pass/fail/gray-area conclusions with price ranges; the four master perspectives challenge each other (for example, Duan Yongping looks at business models, Buffett at financial valuation, Munger at inversion, and Li Lu at long-term certainty), thereby exposing contradictions that a single prompt struggles to produce; it includes built-in anti-bias mechanisms such as information richness A/B/C ratings, a quick rejection checklist, anti-consensus checks, and the principle of leaving blanks; and it requires key data to be cross-validated from at least two independent sources. Skill Composition (20) - Deep research: /investment-research, /investment-team (4 Agents in parallel), /management-deep-dive, /private-company-research, /deep-company-series - Earnings analysis: /earnings-review, /earnings-team - Industry screening: /industry-research, /industry-funnel, /quality-screen, /bottleneck-hunter, /era-alpha, /investment-checklist - Position management: /income-investment, /portfolio-review, /thesis-tracker, /thesis-drift, /news-pulse - Thinking tools: /dyp-ask, /financial-data, /wechat-article Installation and Usage Claude Code users copy skills to the commands directory via scripts/install-claude-commands.sh (or .bat); Codex users generate ~/.codex/skills via install-codex-skills.sh, with an optional slash prompts compatibility layer. The repository maintains three entry sets—skills/*.md, codex-skills/*/SKILL.md, and codex-prompts/*.md—synchronized by scripts. Invocation is by slash command plus company name, for example /investment-research Tencent, /investment-team Meituan, /industry-funnel AI compute. Tools and Engineering Details tools/financial_rigor.py provides financial verification, valuation verification, multi-source cross-validation, three-scenario valuation, Benford's Law detection, and a precise calculator, all using Python decimal.Decimal rather than float. The README also mentions that it can be paired with Claude Code's built-in /deep-research orchestrator to perform multi-Agent adversarial verification of key facts. Repository Contents and Disclaimers The repository contains a large number of sample research reports (the README states 2375 reports, 111 companies, and 23 topics) and shows screenshots of the author's self-reported live trading returns compared with indices. The README explicitly states that the project is for learning and research only, does not constitute investment advice, and that historical returns do not represent future performance. Suitable Audience Individual investors and researchers who want to use an AI client for systematic company research, in-depth earnings reading, industry screening, or position discipline tracking; teams that need a reproducible, structurally consistent investment research output process.