I’m experiencing an issue where Cursor continues to behave as if it were in Ask mode, even though I have explicitly set the prompt/mode to Agent and am requesting code modifications.
Context:
Environment: Windows (c:\lucido\lucidoDevel\gaspi\gaspi2024_www project)
I previously used Ask mode for read‑only explanations.
Now I need Cursor to actually edit files (Agent behavior).
I uninstalled Cursor and did a clean reinstall.
In my latest sessions, I:
Set the mode/prompt to Agent.
Explicitly asked the assistant to “modify these files” and “apply these changes.”
The assistant logically describes the changes and sometimes even shows patch‑like output, but then says it is in Ask mode and must not make edits, so no actual modifications are applied to the files.
Expected behavior:
When the UI or prompt is set to Agent mode and I clearly request changes (e.g. “please modify X.ts like this”), the assistant should be allowed to:
Apply edits to my local files.
Use editing tools (ApplyPatch, etc.) without being blocked by an Ask‑mode restriction.
Actual behavior:
The assistant repeatedly responds with messages like:
“Ask mode is active. I MUST NOT make any edits…”
This happens even after:
Switching mode to Agent in the prompt/settings.
Restarting Cursor.
Fully uninstalling and reinstalling Cursor.
What I’d like to know:
Is there any persistent configuration, project‑level setting, or hidden flag that can force a workspace or conversation into Ask mode regardless of my current mode setting?
How can I completely reset this so that:
Agent mode can actually edit files again in this project, and
The assistant no longer insists it is in Ask mode when I’ve requested Agent behavior?
If needed, I can provide example conversation snippets where:
The UI/prompt indicates Agent mode.
I ask for edits.
The assistant still refuses to modify files due to “Ask mode” constraints.
Hey, thanks for the detailed description. This is an unusual issue. The agent shouldn’t get “stuck” in Ask mode after switching.
A few things that will help us debug:
We need diagnostics:
Help > About, what version are you on?
A screenshot of the agent with the selected mode (when it refuses to edit)
The Request ID from the chat where this happens (three dots in the chat > Copy Request ID)
Try this right away:
Start a new chat CMD/CTRL + I and select Agent mode again
Or open a different folder or workspace to see if the issue persists
Question: Do you see the mode dropdown in the agent UI? Does it show “Agent” or “Ask”?
This looks like a bug where the UI shows one mode, but the model receives instructions for a different one. With your info, we can pass it to engineering to fix.
I can see the screenshot: the UI shows “Agent”, but the model responds like it’s in Ask mode. This looks like a desync bug between the UI and the system prompt that gets sent to the model.
We need a bit more debugging:
Check the .cursorrules file in the root of the gaspi2024_www project. Open it and paste the contents here. It might include something like “you are in ask mode” or “do not edit files” that overrides the mode.
Try in a different workspace or folder:
Open a completely different project
Create a new chat in Agent mode
Ask it to change something
Does the issue happen again?
Try resetting chat settings:
Create a new chat (CMD/CTRL + I)
Pick Agent mode from the dropdown
Don’t switch between modes
Give an edit request right away
If it only happens in this workspace, it’s probably caused by .cursorrules or workspace settings. If it happens everywhere, I’ll pass it to the engineering team with your Request ID.
I’ve noticed an issue while using Ask Mode and it is coming in many accounts used in my organization. Even when Ask Mode is selected, Cursor sometimes enters a loop where it starts analyzing and attempting to implement code, similar to Agent Mode, despite not having access to it.
Interestingly, when I explicitly mention in the prompt that it is not in Agent Mode, the behavior becomes correct and it responds as expected for Ask Mode.
This makes me think there may be some confusion in how the mode context is being interpreted, even when the mode is set correctly in the UI.
Sharing this in case it helps identify and fix the issue. Please let me know if you need any additional details or reproduction steps.
Thanks
Steps to Reproduce
If you told it to create code for a specific module with some context then it start behving like this. This behaviour comes 3-4 times from 10 cases, but when it start behaving like this then it only stop when i tell it that you are in ask mode.
Example Prompt : need to open this mdoal component when user clikc on nav menu locationclick on the location (as you can see in the sceocn detail), do creaet this address selection and adddress pick ui in the d2c-ui and use it into the fs mobile and check for the header component where i will add the click logic to open this modal, This modal will open from the bottom with smooth transition animation, analyze this and create this first with hard coded data, then i will give you api and their response, also when user click on “Use your current location” then take access from teh user lcoatino and fetch their lat and long, so that we can access user exact location, to call our addresss api.
Is this loop analysis consume the tokens consistently, if yes then this must be resolved on priority ?
Yes, when i tell him explicitly then it work fine.
Lately the agent hasn’t been able to determine what mode it’s on and tries to change code on ask or plan mode. I defaulted to always adding “you’re on x mode” at the end of my prompts. Not the end of the world but annoying. It’s a new development that started a couple weeks ago
Steps to Reproduce
Put the agent on ask or plan mode. Ask for it to do something related to changing code. It will attempt to create/change code despite not being allowed to.
Expected Behavior
It should automatically detect what mode it’s on and act accordingly.
This looks like a regression or an incomplete fix. There was a similar issue that was marked as fixed, but you’re on version 2.4.21 and you still see the bug.
Can you try this:
Copy the Request ID from the chat context menu (top right) when the bug happens
Send it here, it’ll help the team understand what’s going wrong
The workaround you’re already using (“you’re on x mode”) is still the only way to get around it for now. I’ll let the team know the bug is still happening on the latest version.
When I ask the cursor agent about a bug in ask-mode it immediately tries to solve it despite not having access to solve it. It gets stuck in an infinite loop of reads until i stop it.
Steps to Reproduce
Ask it to fix sth. but stay in Ask-Mode
Expected Behavior
It should tell me that it doesn’t have the proper access to fix the bug and needs agent mode, or give me the code update as text.
Hey, thanks for the report. This is a known issue. The team fixed it before, but based on your screenshot and a few other reports from the last 2 days, it looks like it’s back.
Can you share the Request ID from this chat? (three dots in the chat > Copy Request ID) This will help confirm it’s a regression and pass the info to the team.
When asked a question about the code, cursor assumes that I am pointing out a bug, and tries to make a change. It even says that it made the change in its reply.
Steps to Reproduce
Ask a question in ‘Ask’ mode. Probably something where you know there’s a bug in your code.
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
Hey, thanks for the report. This looks like a known issue: the agent in Ask mode was trying to edit files instead of staying read-only. The bug was fixed on January 14, but it hasn’t made it into the latest release yet. I think it’ll be available soon.
If the issue is still there after you update, let me know. Please include an example prompt and what the AI is trying to do.
