Codex Desktop exits after 26.609/26.611 updates due to cua_node runtime staging/finalization issue

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

26.609.3341.0（OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g0）

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows11

What issue are you seeing?

Codex Desktop for Windows exits a few seconds after launch on version 26.609.3341.0. The UI never remains open. This started after the update from 26.608.1337.0 to 26.609.3341.0.

Codex CLI still works on the same machine, so this appears to be specific to the Windows Desktop app startup path rather than account authentication or service availability.

Environment:

OS: Windows 11 / Windows NT 10.0.26200
Codex Desktop package: OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g0
Codex app executable:
C:\Program Files\WindowsApps\OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g0\app\Codex.exe
Embedded Chromium version observed in Crashpad metadata:
149.0.7827.54
WebView2 runtime installed:
C:\Program Files (x86)\Microsoft\EdgeWebView\Application\149.0.4022.62
GPU from dump metadata:
Intel GPU, vendor 0x8086, device 0xa7a1, driver 31.0.101.3688

Actual behavior:

The app exits within a few seconds. Windows AppModel logs show the AppX container being created and then destroyed shortly after.

Example AppModel Runtime events:

Process created:
OpenAI.Codex_2p2nqsd0c76g0!App
A few seconds later:
Desktop AppX container destroyed

Codex text logs only contain startup lines, for example:

Launching app agentRunId=null allowDebugMenu=false allowDevtools=false allowInspectElement=false buildFlavor=prod enableSparkle=false enableUpdater=true nodeEnv=undefined packaged=true platform=win32
Appshot hotkey inactive configured=true enabled=true platform=win32

No JavaScript exception or obvious application error is written to the text log.

Crashpad dumps are usually created on normal failed launches. Recent examples:

2026-06-12 13:40:27 36207856 352d6c55-58d9-4217-b205-5332c82d4cd9.dmp
2026-06-12 13:44:53 36178128 2a9c0498-d356-47e2-8fdf-515c78deaffe.dmp
2026-06-12 13:49:45 36173632 9c798ad0-e13f-4856-a1b0-306983c4f06e.dmp

Readable strings in the dump include:

in __char_to_wide: Illegal byte sequence
Illegal byte sequence
\\?\C:\Users\<USER_NON_ASCII>\AppData\Local\OpenAI\Codex\runtimes\cua_node\.staging-...
\\?\C:\Program Files\WindowsApps\OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g0\app\resources\cua_node
...\node_modules\better-sqlite3\build\Release\better_sqlite3.node
...\resources\native\windows-updater.node

The non-ASCII user path appears mojibake-encoded in the dump immediately near Illegal byte sequence / __char_to_wide.

What steps can reproduce the bug?

What steps can reproduce the bug?

Install or update to Codex Desktop 26.609.3341.0 on Windows.
Launch Codex Desktop via either:
codex app
or directly:
C:\Program Files\WindowsApps\OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g0\app\Codex.exe
Observe that the UI appears briefly or does not stay open.
The process exits within a few seconds.
A Crashpad dump is usually written under the Codex Crashpad reports directory.

This is a desktop startup crash, so I do not have a session id, token limit usage, or context window usage for the failing Desktop launch.

What is the expected behavior?

What is the expected behavior?

Codex Desktop should launch and stay open.

Additional information

Regression window:

Windows AppX deployment logs show these Codex package updates:

• 2026-06-05: 26.601.2237.0 -> 26.602.3474.0
• 2026-06-06: 26.602.3474.0 -> 26.602.4764.0
• 2026-06-12 12:19: 26.608.1337.0 -> 26.609.3341.0

The desktop app worked before the recent update and started exiting immediately after the 26.609.3341.0 update.

Additional observations:

• Launching with --disable-gpu still crashes.
• Launching with a fresh --user-data-dir still crashes.
• Launching with an ASCII-only temporary user data dir, e.g. C:\CodexTempProfile, still crashes.
• Temporarily overriding LOCALAPPDATA, APPDATA, and USERPROFILE to ASCII-only paths prevented a Crashpad dump from being written in one test, but the AppX process still exited. This may indicate AppX activation/environment handling is also involved.
• I also tried launching under a separate local Windows account with an ASCII username. The app appeared to exit there as well, although I could not fully inspect that account’s Codex logs from the original account due to Windows permissions.

Suspected cause:

This may be a Windows regression in the cua_node runtime staging / path conversion path active in 26.609.3341.0. The dump strongly suggests a narrow/wide string conversion failure:

in __char_to_wide: Illegal byte sequence

The failure appears while handling paths under:

%LOCALAPPDATA%\OpenAI\Codex\runtimes\cua_node\.staging-...

Requested mitigation:

Could you provide either:

• a patched Codex Desktop build that fixes the Windows cua_node runtime staging/path conversion crash, or
• a download link for the previous working signed Windows package, OpenAI.Codex_26.608.1337.0_x64__2p2nqsd0c76g0, so affected users can temporarily roll back?

The local machine no longer has a usable .msix / .appx installer for 26.608.1337.0. Windows deployment logs show that package was removed and moved under C:\Program Files\WindowsApps\Deleted\..., but it is not available as a clean reinstall source.

Files available for maintainers:

The affected machine has:

• Codex desktop logs:

%LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Local\Codex\Logs\2026\06\12\codex-desktop-*.log

• Crashpad dumps:

%LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Roaming\Codex\web\Codex\Crashpad\reports\*.dmp

I can provide the relevant .dmp and .log files privately if needed.

Additional information

_No response_