VS Code extension: infinite reconnect loop + "stream disconnected before completion"

What version of the IDE extension are you using?

26.5406.31014 (pre-release); also reproduced on 26.406.31014 (stable)

What subscription do you have?

ChatGPT Go

Which IDE are you using?

VS Code

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

The VS Code Codex extension repeatedly enters a reconnect loop and fails to complete responses.

Symptoms:

• "Reconnecting... 1/5 → 5/5" loops continuously
• Eventually fails with:

"stream disconnected before completion"

• Response begins but disconnects mid-stream
• Retry loop does not recover

Key observation:

• This also happens in completely fresh chats / sessions
• So it does NOT appear to be related to large context size or long threads

What I already tried:

• Fresh chat/session
• Forked chat
• Switched between stable and pre-release extension versions
• Issue still persists

Versions tested:

• Pre-release: 26.5406.31014
• Stable: 26.406.31014

Example request ID from one failure:
fd0ab001-6841-4796-8cf8-5fc3d99a8789

What steps can reproduce the bug?

1. Open Codex in VS Code
2. Start a session or open a chat
3. Send a prompt
4. Response begins streaming normally
5. Mid-response, the stream disconnects
6. Extension enters reconnect loop:

• "Reconnecting... 1/5" through "5/5"

1. Retry loop fails and response does not complete

Also reproduced in:

• Completely fresh chats
• Both stable and pre-release extension versions

What is the expected behavior?

Codex responses should stream normally and complete without entering repeated reconnect attempts.

If a temporary connection interruption happens, the extension should either:

• recover cleanly and continue the response, or
• fail once with a clear error

It should not get stuck in a reconnect loop where the response never completes.

Additional information

This was initially suspected to be caused by a very large long-running chat, but the same behavior was reproduced in a completely fresh chat/session.

That seems to rule out context size or my i3-5005U as a low power chip as the root cause.

This issue appears more likely to be related to streaming / reconnect handling in the VS Code extension or backend connection path.