Does not seem to work well when bench commands are involved. For those who mastered parallel development (human or AI!) Any workarounds?
Not sure “mastered” is the right word, but…
Git worktrees never worked well for me with bench commands either. I ended up running multiple benches (one for each project) instead, each in its own Docker container. That keeps bench-specific commands isolated, avoids conflicts, and lets you work on several projects in parallel without them stepping on each other. Much smoother than juggling everything in a single bench.
There’s also a new project that should make this easier: GitHub - frappe/pilot: Frappe Server Manager · GitHub , haven’t tried it yet though.
On the AI side: Me personally, I can’t work on more than 2–3 things at a time anyway, at least for serious engineering work rather than vibe-coded prototyping. I’m also not a fan of spinning up multiple agents in parallel on the same context. Instead I task a frontier model with orchestrating sub-agents and explicitly tell it not to use worktrees and to work sequentially where needed. The resulting code was noticeably more maintainable.
I tried Kilo Code (VSCode extension) which manages worktrees and allows you to then merge them back into your working branch, its nice but of course is limited to “just” coding (not testing, not running server, …).
I worked a bit on running Frappe in Docker containers and exposing OpenCode servers inside these Docker Compose projects. I did not go further even though the technique show promise.