merge conflict resolution

[mikestaub]

What is the overall strategy when agents modify the same file in different worktrees? Should there be a specific agent to handle this?

[nachoal]

It's the same as working with devs, merge the PR that makes more sense, then ask the conflict branch agent to solve the issues with a merge/rebase and give it context so it knows how to rebase it.

Can you share a bit about your workflow? how are you getting these conflicts?

[mikestaub]

Makes sense. These come up naturally as multiple agents work simultaneously. Agent A fixes timing issue with fileA, while agent B refactors the same method for performance for example. I was toying with the idea of creating a 'perforce-like' file checkout system, where all the agents would communicate what files they plan on changing to a global MCP server. If another agent is already mutating that file, maybe the agent switches to another branch or can decide to proceed what it has the context of what other changes are happening.

[nachoal]

I don't think you need all that, that's the point of having git worktrees, each worktree its an isolated copy of the project that doesn't mess up your main repo, every time you run fleet create [name of branch here] you'll have its own isolated environment to work on. Once the agent is done you can simply ask it to generate a PR with the changes and you'll end up with an easy way to merge/fix conflicts

[mikestaub]

I think you are right. The real question then is how does the swarm tackle a backlog? Can you imagine a world where they can autonomously merge 40 PRs over the weekend? At that point, you need a higher-level coordinator that understands the project structure and goals, who can generate the dependency graphs between the tasks.