macOS app: stream disconnected before completion after image_generation_call events
What version of the Codex App are you using (From "About Codex" dialog)?
Observed in local logs as 0.122.0-alpha.13
What subscription do you have?
ChatGPT sign-in
What platform is your computer?
macOS
What issue are you seeing?
Repeated:
stream disconnected before completion: An error occurred while processing your request...
This does not appear to be isolated to a single conversation.
For thread 019db9b2-baa2-7970-92f0-568f1d16d025, the thread itself later retried multiple times and failed around 2026-04-23 02:50-03:03 in America/Los_Angeles.
What steps can reproduce the bug?
In this case the thread was working normally, then the app received websocket events for an image_generation_call, after which the turn retried repeatedly and failed.
Relevant sanitized sequence from local logs:
2026-04-23 02:50:31 response.output_item.added item.type=image_generation_call status=in_progress
2026-04-23 02:50:31 failed to parse ResponseItem from output_item.added
2026-04-23 02:50:32 unhandled responses event: response.image_generation_call.in_progress
2026-04-23 02:50:38 unhandled responses event: response.image_generation_call.generating
2026-04-23 02:51:39 websocket error type=server_error request_id=7b68ddbe-13be-9caa-a7d5-5a71d25fdd82
2026-04-23 02:54:55 stream disconnected - retrying sampling request (1/5 in 216ms)
2026-04-23 02:56:05 stream disconnected - retrying sampling request (2/5 in 401ms)
2026-04-23 02:57:10 stream disconnected - retrying sampling request (3/5 in 768ms)
2026-04-23 03:00:17 stream disconnected - retrying sampling request (4/5 in 1.503s)
2026-04-23 03:03:26 stream disconnected - retrying sampling request (5/5 in 2.912s)
Request IDs involved:
- upstream error pasted into the thread:
173cb339-4dc3-4c8b-8392-11614ebb67a2 - later failure in the same thread:
7b68ddbe-13be-9caa-a7d5-5a71d25fdd82
Separate same-day failures from another thread:
68b83d22-a510-414d-a424-41737234752d22a4bf92-3114-43be-b5bc-20b82a23f87c
What is the expected behavior?
The turn should either continue streaming normally or fail with a handled and clear error state, without repeated disconnect retries and unhandled image-generation events.
Additional information
Possibly related to #19107.
I have also added a follow-up comment with the corrected request ID and the same sanitized excerpt. If helpful, I can provide a larger sanitized excerpt from macOS app logs and ~/.codex/sessions.
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗