Latest Codex repeatedly reconnects in old sessions and cannot start follow-up tasks #26720

Codex app version

Desktop Codex: 26.602.40724 — released Jun 5, 2026

Platform

• macOS Tahoe 26.5.1
• Apple Silicon MacBook Pro
• Pro plan

Related feedback ID

019e967f-24e8-72c2-ac43-0fad22cbcce8

Summary

The latest Codex Desktop build is still frequently showing connection instability. In this case, the issue mainly happened in an existing/old conversation when I tried to send a follow-up task.

Instead of starting the task normally, Codex entered a reconnecting state and stayed there for a long time. This makes the conversation difficult or impossible to continue.

Observed behavior

In an old/existing conversation:

1. I sent a new follow-up task.
2. Codex did not start the task normally.
3. The UI showed repeated reconnect attempts such as:

Reconnecting 4/5
Reconnecting 5/5

4. The visible error was:

stream disconnected before completion: websocket closed by server before response.completed

5. Codex remained in a long reconnecting/waiting state instead of reliably starting or recovering the task.

Expected behavior

Codex should be able to continue an old conversation and start a follow-up task normally. If a temporary websocket/stream disconnect occurs, it should recover quickly and preserve the session state.

At minimum, Codex should:

• reliably start follow-up tasks in old conversations,
• avoid long reconnecting loops,
• recover from websocket disconnects without leaving the task stuck,
• provide a clear retry/resume path if automatic reconnect fails,
• not require the user to abandon the old conversation and start over.

Actual behavior

The old conversation becomes unstable when sending a new task. Codex repeatedly reconnects, waits for a long time, and may fail with websocket closed by server before response.completed before the task can begin properly.

Why this matters

This is a serious reliability problem. Old conversations contain important task history and context. If follow-up tasks in old sessions cannot start reliably because of reconnect loops, users cannot continue long-running work where the context matters.

This also makes the latest version feel unreliable for normal use. A coding agent should not require users to start a new conversation just to avoid connection instability in an existing one.

Request

Please investigate why existing/old conversations can enter long reconnect loops when starting follow-up tasks in the latest Codex Desktop version. The reconnect path should be much faster and more reliable, and users should have a clear one-click retry/resume option when the stream cannot recover automatically.