ai-berkshire turns Claude Code into a value-investing committee

ai-berkshire is a value-investing research framework built entirely on Claude Code, and it's climbing GitHub fast. The idea: instead of one model spitting out a stock opinion, it runs a committee. Sixteen specialized skills handle company research, financial analysis, industry screening, and portfolio management. And the agents argue. One channels Buffett, another Munger, others Duan Yongping and Li Lu, challenging each other across investing philosophies instead of collapsing into a single bland consensus.

The repo claims real trading numbers behind it, 69% in 2024 and 66% year-to-date in 2025 per the author, which you should take with salt. The interesting part is the architecture, not the backtest.

This is a clean example of where Claude Code skills are headed. It's not auto-trading. It doesn't touch your brokerage. It's a research harness that does the slow, structured analysis a junior analyst would, in parallel, with built-in disagreement so you don't get a yes-man. The adversarial multi-agent setup is the trick worth stealing: consensus is where bad investment decisions hide, and forcing the models to fight is a cheap way to surface what a single prompt would smooth over.

Link: https://github.com/xbtlin/ai-berkshire