You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Windows] Codex Desktop crashes on launch in chrome.dll with 0xC0000005 #25376
What version of the Codex App are you using (From “About Codex” dialog)?
148.00.7778.179
What subscription do you have?
Pro x20
What platform is your computer?
Windows 10
What issue are you seeing?
What issue are you seeing?
Codex Desktop on Windows starts and immediately exits. No visible UI appears.
Launching through Start menu / shell:AppsFolder starts a process briefly, then the AppX container is destroyed. Launching Codex.exe manually from the package copy also exits immediately.
ProcDump captures an unhandled access violation:
C0000005.ACCESS_VIOLATION
Visual Studio dump inspection shows the crash in chrome.dll.
No obvious third-party injected DLLs were present in the module list. The module list appears to contain only Codex/Chromium DLLs and Windows system DLLs.
What steps can reproduce the bug?
Starting the app via icon does nothing (app crashes immediately).
You can check whether your system has this environment variable set:
NODE_OPTIONS=--openssl-legacy-provider
If it exists, try deleting it and then see if Codex launches normally.
That variable was probably set for some dev project, but it ends up affecting everything else on the machine — including Codex, which is an Electron app (Electron runs on Node). Electron doesn't allow --openssl-legacy-provider in NODE_OPTIONS, so when it sees that flag it exits immediately, before showing any window. That's the silent "won't launch."
Launching Codex from the Start Menu / shell:AppsFolder\OpenAI.Codex_2p2nqsd0c76g0!App creates the app process briefly.
No visible window appears.
The process exits immediately.
A new Crashpad .dmp is created on every launch under: %LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Roaming\Codex\web\Codex\Crashpad\reports
Checked Microsoft Store/winget update: no newer package available.
Re-registered the AppX package with Add-AppxPackage -DisableDevelopmentMode -Register AppxManifest.xml.
Retested launch: still created a new dump immediately.
Renamed the UI cache folder so Codex recreated a fresh cache.
Retested launch: still created a new dump immediately.
Ran Reset-AppxPackage for OpenAI.Codex.
Retested launch: still created a new dump immediately.
Uninstalled and reinstalled Codex from Microsoft Store via winget.
Retested launch: still created a new dump immediately.
Latest post-reinstall dump was generated at 2026-06-12 20:13:26 Asia/Jakarta. I am not attaching the dump publicly by default, but it is available if maintainers need it through a safer channel.
Follow-up: on the same machine, launching through the Start Menu / shell:AppsFolder\OpenAI.Codex_2p2nqsd0c76g0!App still crashes immediately and creates a Crashpad dump.
However, directly launching the installed package executable works and does not create a new dump:
After direct launch, multiple Codex.exe child processes remain alive, plus the bundled app\resources\codex.exe, and no new Crashpad dump is generated. This suggests the issue on this system is tied to the AppX/Start Menu activation path rather than the bundled app binary being completely unusable.
josemanuelflorezgaviria-max commented on Jul 14, 2026
Crashes 2 and 3 have the exact same minidump signature: 0xC0000005 at chrome.dll+0x2E08F46. In run 3, the Desktop process stayed up for about 10 minutes, then the final log event was Browser Use's embedded webview reaching dom-ready on about:blank; the Crashpad dump followed about 1.3 seconds later. The log ends abruptly, with no normal shutdown sequence.
Sanitized log references for run 3:
lines 124-167: bundled Browser plugin reconciliation; the final config write preserves enabled=false
lines 948-977: embedded IAB/Browser Use lifecycle
lines 955-977: hidden Browser Use webview creation, attachment, and the final dom-ready event
Additional isolation performed:
chat_processes.json contained NUL bytes before run 1. It was quarantined; the associated parsing error disappeared and the file was not regenerated, but crashes 2 and 3 still occurred.
browser@openai-bundled remained enabled=false throughout run 3. Codex reconciled the plugin while preserving that state, but the separate embedded IAB subsystem still created its Browser Use webview and the crash persisted.
No OOM, GPU-process failure, render-process-gone, app-server exit/disconnect, or normal shutdown event appears immediately before the crash.
I did not reinstall because the AppX package is healthy, reinstalling would deploy the same build, and this issue already reports that reinstall did not change the failure.
This appears to rule out the corrupted chat-process state and the Browser plugin enable flag as sufficient explanations. It does not rule out the embedded Browser Use/Electron/Chromium path; the repeated chrome.dll offset immediately after webview readiness makes that path the strongest current correlation.
Would the maintainers like the latest dump, sanitized logs, and the parsed metadata retained from the rotated dumps through a private channel? Also, is 26.707.9981.0 known to contain either a fix or a regression related to this Chromium/IAB crash signature?
Activity
github-actions commentedon May 31, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
leiyan commentedon Jun 6, 2026
You can check whether your system has this environment variable set:
NODE_OPTIONS=--openssl-legacy-provider
If it exists, try deleting it and then see if Codex launches normally.
mosmondor commentedon Jun 6, 2026
Well, I'll be damned - it worked. How? What?
leiyan commentedon Jun 6, 2026
That variable was probably set for some dev project, but it ends up affecting everything else on the machine — including Codex, which is an Electron app (Electron runs on Node). Electron doesn't allow --openssl-legacy-provider in NODE_OPTIONS, so when it sees that flag it exits immediately, before showing any window. That's the silent "won't launch."
mosmondor commentedon Jun 6, 2026
Kudos, really.
ImamHub commentedon Jun 12, 2026
I am seeing the same immediate launch crash on the Microsoft Store Windows app.
Environment
OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g026.609.3341.0OkOpenAI.Codex_2p2nqsd0c76g0!AppBehavior
shell:AppsFolder\OpenAI.Codex_2p2nqsd0c76g0!Appcreates the app process briefly..dmpis created on every launch under:%LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Roaming\Codex\web\Codex\Crashpad\reportsCrash metadata from latest minidump
0xc00000050x7ffb3c5509ab0x0,0x1f0chrome.dll...\OpenAI.Codex_26.609.3341.0_x64__2p2nqsd0c76g0\app\chrome.dllTroubleshooting already tried
Add-AppxPackage -DisableDevelopmentMode -Register AppxManifest.xml.Reset-AppxPackageforOpenAI.Codex.winget.Latest post-reinstall dump was generated at
2026-06-12 20:13:26 Asia/Jakarta. I am not attaching the dump publicly by default, but it is available if maintainers need it through a safer channel.ImamHub commentedon Jun 12, 2026
Follow-up: on the same machine, launching through the Start Menu /
shell:AppsFolder\OpenAI.Codex_2p2nqsd0c76g0!Appstill crashes immediately and creates a Crashpad dump.However, directly launching the installed package executable works and does not create a new dump:
After direct launch, multiple
Codex.exechild processes remain alive, plus the bundledapp\resources\codex.exe, and no new Crashpad dump is generated. This suggests the issue on this system is tied to the AppX/Start Menu activation path rather than the bundled app binary being completely unusable.josemanuelflorezgaviria-max commentedon Jul 14, 2026
I can reproduce the same native Windows crash on a newer Codex Desktop build, with a stable fault offset across two independently captured dumps.
Environment and update
26.707.9981.026.707.9564.0at2026-07-14 08:44:46local timeGet-AppxPackagereportsStatus: Ok; AppX deployment completed successfullyCrash evidence
09:09:30276409:33:042592040,700,704bytesec0eaf1a11d0129c325e71ba25ba61bdbb40ec85ab5e33c1c3d2e50831a88c6509:57:191027640,753,232bytes9b240ec369f3c1e769c04a40255a8494ec0c91dcf1e98e5d0245c5f0f4bea1a2Crashes 2 and 3 have the exact same minidump signature:
0xC0000005atchrome.dll+0x2E08F46. In run 3, the Desktop process stayed up for about 10 minutes, then the final log event was Browser Use's embedded webview reachingdom-readyonabout:blank; the Crashpad dump followed about 1.3 seconds later. The log ends abruptly, with no normal shutdown sequence.Sanitized log references for run 3:
enabled=falsedom-readyeventAdditional isolation performed:
chat_processes.jsoncontained NUL bytes before run 1. It was quarantined; the associated parsing error disappeared and the file was not regenerated, but crashes 2 and 3 still occurred.browser@openai-bundledremainedenabled=falsethroughout run 3. Codex reconciled the plugin while preserving that state, but the separate embedded IAB subsystem still created its Browser Use webview and the crash persisted.render-process-gone, app-server exit/disconnect, or normal shutdown event appears immediately before the crash.This appears to rule out the corrupted chat-process state and the Browser plugin enable flag as sufficient explanations. It does not rule out the embedded Browser Use/Electron/Chromium path; the repeated
chrome.dlloffset immediately after webview readiness makes that path the strongest current correlation.Would the maintainers like the latest dump, sanitized logs, and the parsed metadata retained from the rotated dumps through a private channel? Also, is
26.707.9981.0known to contain either a fix or a regression related to this Chromium/IAB crash signature?