i encountered this situation as well, is it possible to implement a file/function level lock/lease when multiple remote agents are working on the same repo, i discovered this is somewhat doable on local filesys when spinning off multiple openhands-cli. achievable by creating multiple worktrees

1. I found it very useful to tell all of them "There is another agent working on this branch, so be careful, check first, don't overwrite their changes." 2. 'file_reservation' tool(s) to lock and release intently also appear to work, even when the implementation doesn't enforce it (it gives an info message or an error message or it tells it to use force)

agree, but file/function level locking only works on local file system though. i am also thinking about how to make it possible when multiple agents are dispatched on a single repo (for example via opening via openHands UI/github App) remotely, they all first scan the original code base and create PRs against a common repo and assume readonly state before working on it. the more agents you dispatch simultaneously the faster you complete each feature/store but conflicts will significantly increase and not be able to be solved manually. This then becomes a classic distributed problem where resource contention occur🤔. I think this is an interesting task that needs to be solved if we want to leverage the maximum capacity of what code agents can do