ChatGPT Codex: How to stop a started task?

Resolved 💬 12 comments Opened Jun 4, 2025 by Formyselfonly Closed Oct 16, 2025
💡 Likely answer: A maintainer (tibo-openai, collaborator) responded on this thread — see the highlighted reply below.

What version of Codex is running?

Not shown in the interface. The version of Codex in use is not visible to the user. Including this detail in the UI or console logs would be helpful.

Which model were you using?

The specific model is not indicated during task execution. It is assumed to be a variant of GPT-4 or another OpenAI model optimized for code generation and task planning.

What platform is your computer?

Operating System: Windows 10 (64-bit)

Browser: Google Chrome 136.0.0.0

Device: Desktop

Environment: Codex (OpenAI Web IDE)

What steps can reproduce the bug?

1.Open a Codex workspace in Ask mode.

2.Start a task, for example: "Write a unit test for the login function."

3.Once the task is running, attempt to stop or cancel it.

4.Observe that there is no visible Cancel or Stop button.

5.Wait for the task to complete or get stuck, with no option to intervene.

What is the expected behavior?

Users should be able to:
Cancel or stop a running task at any time.
See a clear indication that the task is in progress.
View status updates such as Running, Queued, Cancelled, or Completed.
Avoid being locked into a long-running or incorrect task.

What do you see instead?

No visible stop or cancel control.
No progress indication or timeout handling.
Uncertainty about whether the task is still running or has stalled.
No feedback on whether it's safe to refresh the page or start a new task in the meantime.

Additional information

This issue reduces user control and creates confusion, especially in exploratory or rapid development workflows.
Having the ability to stop a task is important in scenarios where output may take several minutes or becomes irrelevant mid-way.
Suggest adding UI elements for task interruption, status visibility, and timeout warnings.
More than that,software engineering agent in the cloud,so how can OpenAI avoid some hack such as Prompt Hacking?

View original on GitHub ↗

12 Comments

dreasgrech · 1 year ago

Can confirm that when my startup script contained python -m http.server, the task got stuck on that step after launching the server and found no way to cancel it. Moving it to the archive still kept the task running.

The task eventually halted automatically after some time: Startup script timed out after 300 seconds

!Image

Aviel212 · 1 year ago

When an agent is starting to do mistakes it is necessary and convenient to stop it in order to tell him what it was doing wrong, so it can continue working on the task.

There is already a conversation tab, but it is possible to enter a prompt there only when the agent stops. Expected to have there a button that stops the current run, and a text input where you can type while it's working but can't send until you stop it.

gerrywastaken · 1 year ago

I hate it, I make a typo in the prompt which sends the agent off the rails. I realize in the first 5 seconds but must wait till it finishes wasting resources on a pointless task that will never be helpful. We need to be able to stop and edit! I'm shocked they released without even the ability to stop a task. I wonder how many resources have been wasted, totaling over all users, because of this one essential missing feature.

InnerMobius · 1 year ago

yeah seriously we need to be able to stop it. I left out semi crucial info and it's just running all over the place trying to figure something out when I should be able to give it more context.

maxii1996 · 1 year ago

Absolutely. I don't understand why they didn't include a "Stop" button like GPT Chat does. It's obvious that, in the event of errors or whatever, we should be able to stop the execution.

Distortions81 · 1 year ago

Yes, this would be very helpful when it become obvious either the ai agent has misinterpreted what was asked or a simple typo has sent it down the wrong path.

kmnunley · 1 year ago

Seriously this needs to be high priority. I have a task that's been running for a half hour with the agent just attempting to build over and over again and getting stuck. I can't imagine how much resources this is wasting for OpenAI.

metaweta · 1 year ago

+1

hbghlyj · 1 year ago

!Image

Clicking ⏹️ button on the bottom right corner cancels the task

joeflack4 · 1 year ago

Note! The stop button does not appear at the bottom right unless you click into the field!

<img width="907" alt="Image" src="https://github.com/user-attachments/assets/d247fb54-d409-4509-896d-5102f7fbfa9e" />

<img width="1045" alt="Image" src="https://github.com/user-attachments/assets/3a51c7d3-9895-4786-97bb-b6924313fc69" />

gerrywastaken · 1 year ago

Is that new?

tibo-openai collaborator · 9 months ago

This was fixed and we forgot to close, closing now!