stream disconnected before completion reproduces after recreating the conversation

Summary

A conversation that was recreated from scratch still hits the same failure mode during generation. The model starts responding normally, then the stream disconnects before completion and the desktop client retries 5 times.

Steps to reproduce

1. Start a new conversation in the Codex desktop app.
2. Run the same image-generation workflow prompt again in a fresh chat.
3. Let the model begin streaming its response.
4. Observe that the stream disconnects before completion and the client shows repeated reconnect attempts.

Observed behavior

• The response begins normally and then stops mid-stream.
• The client shows:
  • Reconnecting... 1/5
  • Reconnecting... 2/5
  • Reconnecting... 3/5
  • Reconnecting... 4/5
  • Reconnecting... 5/5
  • stream disconnected before completion: An error occurred while processing your request.
• The issue reproduces even after recreating the conversation.

Request IDs

• 3c818545-aead-41f8-a4fe-a75128ed737c
• 8b909f78-0b3c-4d37-96b5-322bbd83d373

Local log signals

• Received turn/started for unknown conversation
• Timed out waiting for structured result

Notes

• I do not see a direct 5xx or socket error in the local logs, but the repeated reconnects and unknown-conversation events suggest a conversation routing or stream stability problem.
• I can share the relevant local log excerpts if helpful.

Environment

• Codex Desktop Version: [请填写版本号，如 26.422.30944]
• macOS Version: [如 macOS 15.x]
• Model: gpt-5.3-codex

Related Issues

This appears to be related to:

• #19440 - Codex App reconnect loop and stream disconnected before completion
• #19690 - Codex Desktop gets stuck reconnecting because failed/completed turns are restored with markedStreaming=true
• #18960 - Frequent reconnect loop in Codex App: websocket closed by server before response.completed