Hi again,
Working with 3.11 in remote-control mode. My normal work requires creating plans, then review, modifications of the plan, then ultimately implementation and then post-implementation review and gap fixes.
However, when working in remote-control the modifications of the plan I request in the Agents Window chat do not land in the chat, in fact I cannot see them. I see the first plan that the agent created (this is mostly true; sometimes they do show the modified plans but I haven’t been able to tell what causes this difference).
I end up asking explicitly to put the modified plan in the local repo’s .cursor/plans and with a specific extension .plan.md so that I can read it in the IDE like a plan and not markdown. Rules I created to this effect have been ignored). This is annoying enough, but then I have to remember to tell the agent not to create a new branch, commit, push and MR every time a plan is modified, which litters the git repo, especially the index, with many unnecessary artifacts.
Rules to try and control this were also ignored because the cloud harness is very explicit about everything being committed and pushed (I see why this is needed for cloud-agents, but not for remote-control). All of this feels like fighting the agent.
I’m not sure what to file this under - a bug (why can’t I see plans after modification)? A feature request (allow no-git ops flows in remote-control)?
Modified plans not showing in the Agents Window. This one is on us, not your setup. When you drive an agent in remote-control, plan updates don’t always sync back to the Agents Window reliably, so you see the original plan and the revisions get dropped. It is something we have seen in this area, and I have flagged your report with the team.
In the meantime, the most reliable way to run your plan → review → revise loop today is in the editor’s Agent panel locally rather than in remote-control:
- Open the Agent panel (Cmd/Ctrl + I) and press Shift + Tab to switch to Plan.
- The plan opens as a virtual file you can read and edit directly, and revisions render inline.
- Click Save to workspace to drop the plan into your repo as a file you can reopen later. That is built in, so you do not need a rule or the
.cursor/plans / .plan.md workaround.
You can then hand the finalized plan off to remote-control for the implementation run. Plan mode has the full flow.
The commit/push on every plan iteration.
In remote-control the agent runs remotely and syncs your changes back through git, so branches and commits pile up. Rules can’t override that, because Rules are guidance for the agent, not a hard control (Rules docs), which is why yours were ignored.
The strongest lever to actually gate the agent’s git commands is a beforeShellExecution hook that denies git commit / git push (Hooks), backed by Auto-Run set to Ask Every Time and removing broad entries like git from the command allowlist (Terminal / Auto-Run). Running the plan → revise loop locally (as above) also sidesteps the churn entirely, since local Plan mode doesn’t push to a remote branch.
Longer term, a lighter, no-git-style remote-control flow is a direction we’re actively working on, and I’ve added your feedback to it.
Thank you @mohitjain , I didn’t realize this thread was answer (I got no push).
Where can I subscribe to to see when this issue has been solved and I can start working normally with the remote-control and not in this convoluted set of workarounds I need to discipline myself not to forget?