came across a self hosted multi agent setup that uses claude code as a runner and the coordination layer is genuinely interesting

been thinking a lot about how claude code fits into broader agent workflows and found a project that uses it as an execution runner within a larger multi agent orchestration system. the project is voice first, macOS only, and self hosted, you say a wake word and a team of agents coordinates the work, with one planning, one executing through claude code, and one reviewing the output before anything ships. what is interesting from an anthropic perspective is how they handle the review loop problem. agents reviewing each other can get stuck in endless cycles and they solved it with a parent child structure plus a watchdog layer which is a cleaner approach than most implementations i have seen. it is v0.1.0 and early so expect rough edges. repo is [github.com/OpenYabby/OpenYabby](https://github.com/OpenYabby/OpenYabby) if you want to see how the claude code integration works within the orchestration.