Show HN: Agentsh – A Decentralized, Scalable Multiagent Harness

Cool! your github shows a demo of 1024 agents working together...I'm actually super curious how this works in practice:

When you have that many agents running, how often they are committing to the repo? do they run in different worktrees, or all in the same worktree and just make small commits with the files they touch?

IF they are running in different worktrees how often do they use the coordination tools across worktres, and is coordination even needed in that case since you have worktree isolation? If they are running in the same worktree, do they coordinate well enough that no agent tries to do things like run a full test suite in the shared worktree? or does that happen and they just deal with it?