Claudex Loop makes Claude Fable and GPT-6 Astra review and build together
An open-source repo with 2k+ stars lets two models challenge each other, alternating between writing code and reviewing it, and cites a study on which review direction helps.
Hey family I found something helpful Free GitHub repo that makes two rival models work as a team. It’s called Claudex Loop. Built by Chase Hanigan. The idea: One model grades itself and says it’s perfect. This repo makes Claude Fable and GPT-6 Astra challenge each other instead. Pinned GitHub: https://github.com/chaseai-yt/claudex-loop How it runs: 1. Fable researches the project and locks a plan 2. Astra reviews the plan 3. They loop until both agree 4. Then they swap Astra builds, Fable audits A July study on 116 coding tasks found the direction matters: Claude reviewing Codex lifted pass rates by 18 points. Codex reviewing Claude made results worse. Repo already has 2k+ stars. Video below 👇 Would you use this for planning, coding, or both?