agentic gitflow?
maybe!
worktree is indeed a good parallel coding scheme to isolate the case of different agents/providers developing the same project in parallel.
However, there are a lot of problems here.
The worktree itself is designed and used by human programmers, and it takes a lot of effort to merge PR and resolve conflicts every time.
The risk of conflict increases dramatically after the Agent works in parallel as an independent coder.
So it is necessary to strictly limit the boundaries in the respective agent.md, not to cross the lightning pool half a step.
Common templates do not allow unauthorized changes by business agents, and separate agents/skills are required for maintenance and PR.
Then it seems that the local needs to have a recording mechanism similar to the shared context, so that the various agents in the development process know each other's status, similar to the human team's irregular state synchronization.
If you really want to do an oversized project of the level of 20 concurrent agents, it is still complicated to think about, but there is a huge opportunity here.
