ai-berkshire

Why we included this project

Ask a chatbot to analyze a stock and you usually get a balanced essay that hedges every claim and ends with a disclaimer. This project is a deliberate alternative: a set of reusable skills that run inside Claude Code or Codex, encoding the value-investing methods of Buffett, Munger, Duan Yongping, and Li Lu so a single person can drive a multi-agent research workflow. The framework forces a concrete verdict (pass, fail, or gray zone) with price ranges, runs four independent analyst agents in parallel so genuine disagreements surface, and includes anti-bias checks like an information-richness rating and a quick-kill veto list. It also handles the practical details that trip up LLM analysis, such as verifying market caps with exact decimal arithmetic and cross-checking key figures across sources. For individual investors, analysts, and small teams who want reproducible, decision-oriented equity research, this is a practical way to get past the hedging that makes plain chat prompts useless for decisions.