factory-worktree

Per-agent git worktrees prevent file conflicts during parallel development by giving each agent its own isolated working copy while sharing the underlying repository, keeping changes independent until merged.

Using git worktrees or jj workspaces is the best way to run independent experiments without merge conflicts. Each agent gets a full copy of every file, allowing isolated feature work, independent testing, and clean rollbacks before an atomic merge.

Yes, the parallel development workflow supports both jj workspaces and git worktrees. It handles per-worktree setup, dependency installation, and merge or cleanup workflows using either version control system to unify results.

To merge results after parallel tasks finish, you perform an atomic merge using jj or git to combine changes from dedicated worktrees. You unify results after all agents complete, and discard any worktree where an agent failed.

Yes, each agent can run its own dev server, tests, and linter inside an isolated worktree. Full toolchain access is maintained because every worktree is a complete copy of the repository, enabling independent testing during parallel development.

You should not use git worktrees for multi-agent coding if you need simultaneous file edits within a single shared directory without independent copies. Worktrees are also less suitable if your repository setup lacks support for jj workspaces or git worktree isolation.