You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[FEATURE]: Plan Mode + Question tool can auto switch to Build mode #7801
I have verified this feature I'm about to request hasn't been suggested before.
Describe the enhancement you want to request
Plan mode is ready, it asked me to switch the plan mode
I choose accept/approve see image above but the mode isn't auto switch.
Reacted by Art Shendrik, 0xAuwtch, a2k, Jérôme Burkard, Raid Abderrezak Ouahioune, Daniel Puglisi, Alberto Ortiz, reo, WuChaoli, Alvaro Tanarro and 25 more
A visible Plan/Build mode indicator is not optional — it is a security requirement.
When managing production infrastructure, I need to be absolutely certain that the model is in Plan (read-only) mode before I ask any question. The problem is twofold:
Models make unsolicited changes. Even when you did not ask for modifications, models can and do misinterpret queries and silently modify files. Without a clear indicator, you only realize what happened when it is too late — a config file was rewritten, a service was restarted, a production site went down.
There must be a manual toggle to enforce read-only. The ability to explicitly lock the model into Plan mode (no writes, no edits, no bash) is critical for safe operation. If the only way to guarantee no changes is to trust the model judgment, that is not acceptable for production environments.
The UI must show:
A clear, visible badge/indicator of the current mode (Plan/Build) at all times
A clickable button to toggle between modes (not just a keyboard shortcut that users may not know)
This is not a convenience feature — it is a safety control. Without it, every interaction carries the risk of unintended modifications to production systems.
For context, the keyboard shortcut (Ctrl+. or Tab) is not discoverable, and after the recent desktop update the visual indicator was removed entirely, making it impossible to tell which mode is active at a glance.
Also hitting this with the Question tool specifically: when the AI asks a confirmation question (e.g. "Proceed with these changes?") while in plan mode, confirming doesn't switch to build. I then have to manually switch to build mode, and the AI ends up asking the same/equivalent confirmation again before actually executing — so the question gets asked twice, which wastes tokens on top of the workflow friction.
+1 on auto-switching to build once a plan is confirmed.
So it might be due to the fact that in opencode, it's not actually there are plan "mode" or build "mode" but it's just "plan" agent and "build" agent unlike other harnesses.
After I realized that, I embraced a new agent that is hybrid of both called "plan-build" and I'm pretty happy with how it handle planning and implementing and switches the "mode" seamlessly which is closest that I got from some other harness that provided the auto mode switching experience.
Very often I can already tell the plan is viable halfway through the response, but I still have to wait for the entire plan to finish generating before I can act on it. If the agent could proactively ask "Ready to switch to Build mode?" mid-generation, that would save a significant amount of tokens.
Even without proactive asking, a dedicated shortcut key to approve the plan would be a huge improvement. Right now the workflow is: manually switch to Build mode → type "go" / "yes" / "execute" — which feels unnecessarily clumsy.
Plan mode is essentially generating a TODO list. What if there was a clear, explicit UI action (like a button or shortcut) to approve the plan and start executing the TODOs, instead of having to express approval in natural language every time?
Proposal opened in #49218: question.reply takes an optional agent, and the question tool hands the session over with the same synthetic-message mechanism as plan_exit (subagent sessions are ignored). The TUI shows ctrl+b ("answer & continue in build") when the plan agent asked the question, and the app/desktop question dock gets an equivalent button. Happy to adjust the design here before further review.
Activity
github-actions commentedon Jan 11, 2026
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if your specific case differs from these.
gasatrya commentedon Jan 11, 2026
manifestcowboy commentedon Mar 24, 2026
Need for true agentic workflow
objarni commentedon Mar 30, 2026
Any updates on this, or do we still need the EXPERIMENTAL... env var?
mlopezcoria commentedon Jun 5, 2026
A visible Plan/Build mode indicator is not optional — it is a security requirement.
When managing production infrastructure, I need to be absolutely certain that the model is in Plan (read-only) mode before I ask any question. The problem is twofold:
Models make unsolicited changes. Even when you did not ask for modifications, models can and do misinterpret queries and silently modify files. Without a clear indicator, you only realize what happened when it is too late — a config file was rewritten, a service was restarted, a production site went down.
There must be a manual toggle to enforce read-only. The ability to explicitly lock the model into Plan mode (no writes, no edits, no bash) is critical for safe operation. If the only way to guarantee no changes is to trust the model judgment, that is not acceptable for production environments.
The UI must show:
This is not a convenience feature — it is a safety control. Without it, every interaction carries the risk of unintended modifications to production systems.
For context, the keyboard shortcut (Ctrl+. or Tab) is not discoverable, and after the recent desktop update the visual indicator was removed entirely, making it impossible to tell which mode is active at a glance.
emirBytes commentedon Jul 5, 2026
Feature is so much needed. It would greatly reduce bandwidth usage. January and now July still not added?
jrruethe commentedon Jul 16, 2026
I need this feature for Plannotator.
camiicode commentedon Jul 19, 2026
Also hitting this with the Question tool specifically: when the AI asks a confirmation question (e.g. "Proceed with these changes?") while in plan mode, confirming doesn't switch to build. I then have to manually switch to build mode, and the AI ends up asking the same/equivalent confirmation again before actually executing — so the question gets asked twice, which wastes tokens on top of the workflow friction.
+1 on auto-switching to build once a plan is confirmed.
assawalhy commentedon Aug 14, 2026
What is the progress for this? we are in the age of agentic coding and this simple crucial feature is still not implemented :(
ryuheechul commentedon Aug 15, 2026
So it might be due to the fact that in opencode, it's not actually there are plan "mode" or build "mode" but it's just "plan" agent and "build" agent unlike other harnesses.
After I realized that, I embraced a new agent that is hybrid of both called "plan-build" and I'm pretty happy with how it handle planning and implementing and switches the "mode" seamlessly which is closest that I got from some other harness that provided the auto mode switching experience.
iZerat commentedon Aug 17, 2026
Very often I can already tell the plan is viable halfway through the response, but I still have to wait for the entire plan to finish generating before I can act on it. If the agent could proactively ask "Ready to switch to Build mode?" mid-generation, that would save a significant amount of tokens.
Even without proactive asking, a dedicated shortcut key to approve the plan would be a huge improvement. Right now the workflow is: manually switch to Build mode → type "go" / "yes" / "execute" — which feels unnecessarily clumsy.
Plan mode is essentially generating a TODO list. What if there was a clear, explicit UI action (like a button or shortcut) to approve the plan and start executing the TODOs, instead of having to express approval in natural language every time?
GanzellaLorenzo commentedon Sep 15, 2026
Proposal opened in #49218:
question.replytakes an optionalagent, and the question tool hands the session over with the same synthetic-message mechanism asplan_exit(subagent sessions are ignored). The TUI showsctrl+b("answer & continue in build") when the plan agent asked the question, and the app/desktop question dock gets an equivalent button. Happy to adjust the design here before further review.