/permissions needs to be able to be invoked during tasks

Resolved 💬 9 comments Opened Apr 22, 2026 by gusholcomb Closed Jun 20, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What variant of Codex are you using?

cli

What feature would you like to see?

The only time I want to use /permissions is during an active effort and having to cancel a task just to make it so I can edit the permissions is very frustrating. Please let me invoke it at any point and if you want to not apply it until after the current task runs thats fine and it can be queued up, etc.

It is a massive inconvenience to not be able to upgrade permissions when I'm confided the agent is on the right path, etc.

Additional information

_No response_

View original on GitHub ↗

9 Comments

github-actions[bot] contributor · 2 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #18854
  • #17653

Powered by Codex Action

gusholcomb · 2 months ago

not a duplicate

Mahkhmood9 · 2 months ago

I assume this is intentional by design, to prevent confusion.
Example of the confusion: Why does it keep asking for permissions when they've already been set?

The key thing to understand is that any permission changes will only take effect on the next prompt.

I encounter the same behavior in the terminal — however, in VS Code, permissions can be changed mid-session, with the understanding that they apply from the next prompt onward.

(Interestingly, one could argue this is justified "_by design_," given the nature of the UI and the async architecture inherent to Electron-based desktop applications.)

gusholcomb · 2 months ago

I hear that but at least let me queue up the permission change in the same way it queues up my next prompt

Mahkhmood9 · 2 months ago
I hear that but at least let me queue up the permission change in the same way it queues up my next prompt

you mean "changes will apply next prompt" kind of thing?

gusholcomb · 2 months ago

yes. it's already a built in mechanism:

<img width="683" height="57" alt="Image" src="https://github.com/user-attachments/assets/0b77dbbc-80eb-4c9f-a6cc-1fcbb3478a32" />

Mahkhmood9 · 2 months ago
yes. it's already a build in mechanism: <img alt="Image" width="683" height="57" src="https://private-user-images.githubusercontent.com/1017856/582315738-0b77dbbc-80eb-4c9f-a6cc-1fcbb3478a32.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NzY5MTY2NzEsIm5iZiI6MTc3NjkxNjM3MSwicGF0aCI6Ii8xMDE3ODU2LzU4MjMxNTczOC0wYjc3ZGJiYy04MGViLTRjOWYtYTZjYy0xZmNiYjM0NzhhMzIucG5nP1gtQW16LUFsZ29yaXRobT1BV1M0LUhNQUMtU0hBMjU2JlgtQW16LUNyZWRlbnRpYWw9QUtJQVZDT0RZTFNBNTNQUUs0WkElMkYyMDI2MDQyMyUyRnVzLWVhc3QtMSUyRnMzJTJGYXdzNF9yZXF1ZXN0JlgtQW16LURhdGU9MjAyNjA0MjNUMDM1MjUxWiZYLUFtei1FeHBpcmVzPTMwMCZYLUFtei1TaWduYXR1cmU9NzA4NGIyMjM0ZWNlZGRiNTY4YmJkZTJhMGViNGFhYmVlNTRjZDM5NmY2ODE4Njg2Njg2MWU4NTVhYzhlMDQ0ZCZYLUFtei1TaWduZWRIZWFkZXJzPWhvc3QmcmVzcG9uc2UtY29udGVudC10eXBlPWltYWdlJTJGcG5nIn0.ZVlF0LlRsysjaRTYxvcdNHvGJJ9PUtjwScSXkOvR4Jo">

usally its my job to fail on basic english lol.

I meant like ...

you apply to "Full access" in the "/premission" and then outlined in red , "this wil apply next prompt you sent"

Augtons · 1 month ago

+1 from me.

I think runtime mode switching would be a huge quality-of-life improvement for Codex TUI.

A common scenario is that I ask the agent to make changes, but halfway through I realize its approach is drifting away from what I had in mind. At that point, I'd like to switch to Plan Mode immediately and steer it back on track. Currently, I have to interrupt the session with Ctrl+C first, which breaks the flow.

The opposite situation happens too: when the agent is doing a good job and I gain confidence in its changes, I'd like to switch to a more automated "Auto-review" mode without stopping its current work.

<img width="868" height="261" alt="Image" src="https://github.com/user-attachments/assets/c4b3adba-e929-4674-b0ae-70bf46f585a2" />

Claude Code has supported these kinds of workflow transitions for a while, and they feel very natural in day-to-day use. I'd be excited to see similar capabilities added to Codex TUI.

Mahkhmood9 · 1 month ago
+1 from me. I think runtime mode switching would be a huge quality-of-life improvement for Codex TUI. A common scenario is that I ask the agent to make changes, but halfway through I realize its approach is drifting away from what I had in mind. At that point, I'd like to switch to Plan Mode immediately and steer it back on track. Currently, I have to interrupt the session with Ctrl+C first, which breaks the flow. The opposite situation happens too: when the agent is doing a good job and I gain confidence in its changes, I'd like to switch to a more automated "Auto-review" mode without stopping its current work. <img alt="Image" width="868" height="261" src="https://private-user-images.githubusercontent.com/79037928/601671802-c4b3adba-e929-4674-b0ae-70bf46f585a2.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.3i-hUnX2tWFPs8bvPYIJduNvy6BDTXksSStrwxGTjcg"> Claude Code has supported these kinds of workflow transitions for a while, and they feel very natural in day-to-day use. I'd be excited to see similar capabilities added to Codex TUI.

dude , you need to cancel the current running taks- you can use the esc button too--- as it is saying - there is a task in progress - are you unable to interrupt the task?