Desktop app 26.609.9530.0 crashes on launch — no window ever opens (worked on 26.602.9276.0)

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

26.609.9530.0 (MSIX package: OpenAI.Codex_26.609.9530.0_x64__2p2nqsd0c76g0)

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Windows 10 Pro (build 10.0.19045), 64-bit (x64 / AMD64), locale ko-KR

What issue are you seeing?

After the desktop app auto-updated from 26.602.9276.0 to 26.609.9530.0, the GUI no
longer launches. Starting it (Start menu, or codex app) spawns the Electron child
processes (gpu-process, network/storage service, crashpad-handler), which all exit
within ~1-2 seconds. No main window ever appears. No crash dump is written to Crashpad
and nothing shows up in Windows Event Viewer / WER.

The desktop log (%LOCALAPPDATA%\Codex\Logs\...) is identical for every launch — exactly
two lines, then it stops:

info Launching app ... buildFlavor=prod packaged=true platform=win32
info Appshot hotkey inactive configured=true enabled=true platform=win32

So it dies in the main process right after the "Appshot hotkey inactive" step, before
creating the BrowserWindow / before connecting to the app-server. Version 26.602.9276.0
worked fine (its logs show windows opening and the app-server connecting). This is a
regression introduced in 26.609.9530.0.

Things I already tried — all reproduce the exact same 2-line crash:

• Clean uninstall + reinstall from the Microsoft Store (reinstalls the same 26.609.9530.0)
• Re-register the MSIX package (Add-AppxPackage -Register)
• Launch with a fresh Electron profile (moved %APPDATA%\Codex aside)
• Removed ~/.codex/.codex-global-state.json
• Cleared the cua_node runtime staging folders
• Reboot

The CLI backend is fully healthy (codex doctor = 16 ok, ChatGPT auth valid, websocket
connected). Only the desktop GUI is broken.

Possibly related: the cua_node (Windows Computer Use helper) runtime never finishes
installing — %LOCALAPPDATA%\OpenAI\Codex\runtimes\cua_node accumulates EMPTY .staging-*
folders on every launch, and the 26.602 logs repeatedly show:
warning [computer-use-native-pipe] computer-use native pipe startup failed
errorMessage="Windows Computer Use helper paths are unavailable"
In 26.602 this was a non-fatal warning (status=skipped); in 26.609 the app may now be
treating this early-init step as fatal, which would explain dying right after
"Appshot hotkey inactive".

What steps can reproduce the bug?

1. Be on Windows 10/11 with the Codex desktop app updated to 26.609.9530.0

(in my case it auto-updated from the working 26.602.9276.0 via the Microsoft Store).

1. Launch the app — from the Start menu, OR from a terminal with codex app.
2. Watch the process list: Codex.exe child processes start (gpu-process, network/storage

service, crashpad-handler) and then all exit within ~1-2 seconds.

1. No main window ever appears; the app is simply not running afterward.
2. Open the newest log at %LOCALAPPDATA%\Codex\Logs\<yyyy>\<mm>\<dd>\codex-desktop-*.log —

it contains only these two lines and nothing else:
info Launching app ... buildFlavor=prod packaged=true platform=win32
info Appshot hotkey inactive configured=true enabled=true platform=win32

100% reproducible — happens on every launch. Reinstalling, re-registering, resetting the
Electron profile, removing ~/.codex/.codex-global-state.json, clearing the cua_node
staging folders, and rebooting all produce the identical 2-line crash.

What is the expected behavior?

Launching the app should open the Codex main window (as it did on 26.602.9276.0): the Electron child processes stay alive, the app connects to the app-server, and the UI loads with the thread list / workspace. The "Windows Computer Use helper paths are unavailable" condition should remain a non-fatal warning and must not prevent the main window from opening.

Additional information

_No response_