Inconsistent Plan Mode Output: Missing Interactive Execution Prompt

[BobbyWang0120]

Bug Description I'm reporting an issue with inconsistent behavior in plan mode.

Expected Behavior: When using plan mode, the final output should always be a structured, executable plan that presents the user with an interactive prompt for approval (e.g., a Yes/No choice). This should be the consistent behavior regardless of the request.

Actual Behavior: Sometimes, instead of generating the actionable plan prompt, Claude Code simply describes the intended plan as plain text within its conversational response. The user can read the steps of the plan, but there is no prompt to approve and execute them directly.

Impact: This inconsistency significantly diminishes the user experience of plan mode. It forces the user to guess whether they will receive an actionable plan or just a text description. When the latter happens, it breaks the workflow, as the user then has to manually copy the steps and re-prompt, which defeats the entire purpose of having an automated, approvable planning feature.

A consistent and reliable output format is crucial for plan mode to be effective.

Thanks for looking into this.

Environment Info

• Platform: darwin
• Terminal: iTerm.app
• Version: 1.0.24
• Feedback ID: 9fde3d4d-10ca-4633-8cad-dd11e9b43f18

[kylesnowschwartz]

I second this feature request.

Planning mode as it originally was implemented was preferable to the current feature. Originally you toggled planning mode, and when you were done planning, you toggled it back to regular or accept-all-edits modes. This allowed you to toggle back to regular mode, and then ask it to make a document with your plan.

The current implementation has UX flaws. When the planning task is finished, you are prompted to either:

1. Keep Planning
2. Stop planning If you choose to stop planning, you are automatically put into accept-all-edits mode and Claude just goes brrrrrr with changes. You then have to manually prevent it from doing so to continue planning/implementing in stages and with control. If you're not quick, you get instant changes (including bash commands, very dangerous).

I would suggest you put that power back in the User's hands and allow them to decided what mode they should be in at all times.

[github-actions[bot]]

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.