I run one long-lived Cursor Project as the coordinator for my repo. It looks like the coordinator keeps the same machine for the whole Project, and its checkout doesn’t update when PRs are merged to main. Only an explicit git pull or fast-forward updates it.
So the coordinator works from old files. For example, a rules file already merged to main was missing from its checkout until we fast-forwarded by hand. Our workaround now is to fast-forward main in the environment’s install/start steps (.cursor/environment.json).
Questions:
Is it expected that a Project’s coordinator never refreshes its checkout after merges?
Is there a supported way to keep it in sync with main, e.g. an automatic fast-forward when the checkout is clean and on main?
If not, is fast-forwarding in the environment install/start steps the recommended approach?
Yes, that’s expected today. The coordinator’s checkout is set when its machine starts and doesn’t update on later messages.
There’s no built-in auto-sync today. The easiest fix is to ask the coordinator to make it a habit, for example: “At the start of each turn, if you’re on main with no local changes, run git fetch origin main && git merge --ff-only origin/main.” Ask it to save that to the Project’s shared context so it sticks.
Fast-forwarding in install/start won’t help the running coordinator, since those only run when a machine is set up. For the agents it starts, open your environment in the Cloud Agents dashboard, turn on Update stale builds under Build settings, and set Staleness threshold to 0 (details).
Your auto fast-forward idea makes sense, and we’ll be tracking this thread as feedback