Codex becomes unusable after forced re-login: single instance works, other instances fail, reboot breaks all IDE instances

[Enthoes1]

What version of Codex is running?

0.4.62

What subscription do you have?

chatgpt plus

Which model were you using?

gpt5.2

What platform is your computer?

windows wsl

What terminal emulator and version are you using (if applicable)?

No response

What issue are you seeing?

Description

Today Codex unexpectedly prompted me to log out and re-login.
The login flow itself works normally: I can successfully log out and log back in with my ChatGPT account.

However, after this forced re-login, Codex behavior becomes inconsistent and eventually completely unusable.

Observed Behavior

1. After re-login:

   • One existing IDE instance can still use Codex normally.
   • All other IDE instances immediately fail with:

     stream disconnected before completion:
     error sending request for url (https://chatgpt.com/backend-api/codex/responses)
     

2. Re-login does NOT fix the failing instances.

   • Logging out and logging back in works every time.
   • Authentication appears successful.
   • But Codex requests still fail in those instances.

3. After restarting the device:

   • Codex becomes completely unusable in ALL IDE instances.
   • Every Codex request fails with the same stream disconnected error.
   • Login / logout still works normally, but Codex cannot function anymore.

Key Point

Authentication is successful and stable, but Codex functionality depends on instance state.
After a forced re-login, only one existing instance may continue working temporarily.
Once the system is restarted, all instances become unusable.

This suggests a session / backend / connection handling issue rather than an authentication failure.

Expected Behavior

• After a successful login, Codex should work consistently across IDE instances.
• Restarting the device should not permanently break Codex functionality.
• If the backend connection is blocked or invalid, Codex should fail gracefully or recover after re-login.

Actual Behavior

• Codex works only in one existing instance after re-login.
• Other instances fail immediately.
• After reboot, Codex fails in all instances permanently (until now).
• Login/logout always succeeds, giving no indication of failure.

Error Message

What steps can reproduce the bug?

have mentiened before

What is the expected behavior?

No response

Additional information

No response

[Enthoes1]

Update: I believe I’ve identified a strong clue related to the root cause.

After further investigation, I noticed that the Codex IDE extension auto-updated a few hours before the issue started.
I manually rolled back the Codex extension to a version from ~2 weeks ago, and Codex immediately became usable again.

With the older extension version:

• Codex works normally in the same environment
• No changes to network, proxy, account login, or system settings
• Same machine, same IDE, same ChatGPT account

With the latest extension version:

• Codex fails consistently with:
  stream disconnected before completion
error sending request for url (https://chatgpt.com/backend-api/codex/responses)

This strongly suggests that a recent extension update introduced a regression affecting certain network/proxy environments (possibly related to how sessions, connections, or backend-api requests are handled).

This also explains why:

• Codex prompted a re-login earlier today
• One existing instance kept working temporarily
• After restart, all instances failed

Rolling back the extension fully restores functionality.

Hope this helps narrow down the debugging scope.
Happy to provide exact extension versions or test further if needed.

[etraut-openai]

When you see the "stream disconnected" error, please, use the /feedback slash command to upload your logs and session details, then post the thread ID here.

[etraut-openai]

You mentioned that the problem went away when you reinstalled an older version. Do you know precisely which version the problem started to occur? If so, this would be really useful information for us.

[Enthoes1]

0.4.62, for cursur extension. I find that this version published in 1.22. I never meet this error before. And now I backward to 0.4.58, the error dispear.

[etraut-openai]

I think we've identified the cause of this. We introduced a regression in the TLS handling that causes network failures in certain enterprise network environments. We've reverted the change that introduced the regression. Please try the latest prerelease version of the extension (0.5.66) and let us know if that fixes the problem for you.

[jasminelee]

I'm also having the same issue, no matte the extension version: 0.4.58, 0.4.66, 0.5.58, and 0.5.66.

[Enthoes1]

I think we've identified the cause of this. We introduced a regression in the TLS handling that causes network failures in certain enterprise network environments. We've reverted the change that introduced the regression. Please try the latest prerelease version of the extension (0.5.66) and let us know if that fixes the problem for you.

Thank you for your work; the error was resolved in version 0.4.66