▌ GitHub radar
architect-loop: Claude plans, Codex builds, repo remembers
A Claude Code skill that splits AI work into two roles — one model designs and checks, another writes the code — and uses the Git repo itself as shared memory. It hit hundreds of stars in days.
architect-loop is a Claude Code skill that coordinates two AI models with different strengths: Claude acts as the architect that plans slices of work, locks in acceptance criteria and reviews the result, while a Codex builder does the implementation in its own isolated Git worktree. The architect runs the checks itself instead of trusting the builder's report, and the Git repo holds the shared state so handoffs stay lightweight. There is also a research mode where a scout maps the problem and parallel researchers gather sources under strict limits.
Why a vibe-coder should care
If you already pay for Claude and a Codex-style assistant, this stitches them into one disciplined workflow where work gets verified before it merges. It is a concrete recipe for letting agents tackle bigger tasks without quietly going off the rails.
More finds