In Cursor, Plan mode states that it should be able to create and update plans, but in practice it refuses to edit the .plan.md plan file it just created. I had to switch back to Agent mode to update the plan, which seems like a mismatch between the documented behavior and what’s allowed.
Steps to Reproduce
Steps to reproduce:
Start a conversation and ask the assistant to create a plan (this generates a .plan.md file under .cursor/plans in my user directory).
While still in Plan mode, ask the assistant to update that plan (e.g., change the high‑level approach and todos).
The assistant replies that in Plan mode it cannot modify files or edit the plan, and only gives instructions for manual edits.
Switch to Agent mode and make the same request.
The assistant can now successfully edit the .plan.md file.
Expected Behavior
Expected behavior:
Plan mode should be allowed to edit its own plan files (.plan.md), since its purpose is to collaboratively refine and update plans before execution. The behavior description for Plan mode explicitly suggests it can update plans.
Actual behavior:
In Plan mode, the assistant refuses to update the plan file and instead only describes how I could edit it myself.
In Agent mode, the assistant is able to modify the same plan file directly.
Hey, thanks for the detailed report. This is a known issue. Plan mode sometimes loses the ability to edit its own .plan.md files and says it can’t modify files.
Open a new chat CMD + I, switch back to Plan mode, and try updating the plan again.
Try saying this explicitly in your prompt: “Edit the plan file directly, update the .plan.md” It sometimes helps the model get past the limitation.
Like you noticed, switching to Agent mode to edit the plan works as a fallback.
For debugging, can you share the Request ID from the chat where Plan mode refused to edit? (three dots in the top right of the chat → Copy Request ID). This will help the team narrow down the issue.
The team is aware of the bug. There’s no timeline yet, but your report helps with prioritization. Let me know if the workarounds help.
In plan mode, the agent will create an initial plan, but then any conversation after that which should result in updates to the plan markdown file generally result in the agent stating that since it is in ‘plan’ mode, it cannot update files (including the plan file) and it instead describes the updates to me in the agent chat instead of manipulating the plan.md file. If I argue with it, I can eventually get it to realize that it can, in fact, update the plan markdown file, but this can take two or three prompts (and the associated wasted tokens) to get the agent to realize that it is actually allowed to manipulate plan.md files when in plan-mode. This is a recent regression, but I’m not certain which version introduced this bug.
Steps to Reproduce
Go into plan mode. Ask the agent to produce a plan for something. Then, while still in plan mode, ask the agent to update or modify the plan in some way (for Step 2, please do X instead of Y). Watch the agent generate a bunch of text in the chat window describing how it will update the plan, without it actually manipulating the markdown file. Pay close attention to the “reasoning” as it flashes by, you’ll see it make comments to itself about being read-only due to plan mode.
Expected Behavior
When discussing a plan with the agent in plan mode, the plan markdown file should be updated by the agent as the plan evolves.
Operating System
Windows 10/11
Linux
Version Information
v2.6.13 (user setup)
For AI issues: which model did you use?
No idea, in “auto” mode.
For AI issues: add Request ID with privacy disabled
