codex - 💡(How to fix) Fix Codex App thread cannot continue: remote compact task fails with unknown variant `auto`

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.519.81530 (3178) Bundled agent version: codex-cli 0.133.0

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 25.4.0 arm64 arm

What issue are you seeing?

A long-running Codex App thread can no longer continue. When I send a new message in that existing thread, Codex enters automatic context compaction and then fails before the model can respond.

Full error:

Error running remote compact task: stream disconnected before completion: unknown variant auto, expected high or original at line 1 column 2199103

The issue repeats every time I try to continue the same thread. The thread is now effectively unusable from the Codex App UI.

Environment:

• Codex App version: Version 26.519.81530 (3178)
• Bundled agent version: codex-cli 0.133.0
• macOS Apple Silicon arm64
• Subscription: ChatGPT Pro
• Installed/reinstalled from the official macOS Apple Silicon Codex App DMG

I reinstalled the Codex App from the official Apple Silicon DMG, but the bundled agent still reports codex-cli 0.133.0.

What steps can reproduce the bug?

1. Open an existing long-running Codex App thread with substantial prior context.
2. Send a new message in that same thread.
3. Codex begins automatic context compaction.
4. The run fails before producing a model response.
5. Repeating the same action in the same thread produces the same error:

Error running remote compact task: stream disconnected before completion: unknown variant auto, expected high or original at line 1 column 2199103

I tried:

• Quitting and restarting Codex App
• Killing Codex-related processes
• Reinstalling the official macOS Apple Silicon Codex App
• Reopening the same thread
• Sending a short follow-up message

The same thread still fails during automatic remote compaction.

What is the expected behavior?

Codex should successfully compact the existing thread context and continue the conversation.

If compaction fails, the app should recover gracefully, produce a compact checkpoint/summary, or allow the user to continue the thread without permanently wedging it.

A schema/version mismatch in the remote compaction task should not make an existing long-running thread unusable.

Additional information

I submitted feedback from the broken Codex thread.

Feedback ID: 019c6c45-2709-70b0-afac-7bc572e4451f

The app log also showed a feedback upload requestId: c5d24fcc-3498-41ee-ab79-0a30a1b8c0ba

Local transcript extraction succeeded, which suggests the local transcript itself is readable.

Local transcript extraction results:

• JSON parse bad lines: 0
• Extracted messages: 683
• No "detail":"auto" fields found
• No "image_detail":"auto" fields found

This suggests the failure may be occurring in remote compaction or server-side handling rather than from an obviously corrupted local transcript field.

I am not including project names, repository names, file paths, source code, transcripts, or full logs publicly because they may contain private information.