ChatGPT/Codex App 26.715.4045.0 exits silently seconds after launch after a clean reset; Beta works
What version of the Codex App are you using (From “About Codex” dialog)?
Microsoft Store package: OpenAI.Codex 26.715.4045.0 x64 Internal desktop release from logs: 26.715.31925 (prod) Bundled app-server: 0.145.0-alpha.18 The About dialog is unavailable because the app exits seconds after launch.
What subscription do you have?
ChatGPT Plus
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What issue are you seeing?
The production Microsoft Store app starts, briefly creates its main window, and then the entire AppX process tree disappears about 6 seconds after activation. No error dialog is shown.
Upgrade history and how I reached this state:
- The problem originally started on the previous production package,
OpenAI.Codex 26.707.9981.0. That version produced repeated Windows Application Error Event ID 1000 entries with exception0xc06d007f, associated with the bundledserialport.nodenative module. The production app became unusable because it repeatedly crashed or closed. - I installed and switched to the Codex Beta package as a workaround. Beta remained usable on the same Windows installation and account.
- After openai/codex#33375 was closed as completed on July 17, 2026, I installed the latest production ChatGPT/Codex Store package,
OpenAI.Codex 26.715.4045.0, expecting the original native-module defect to be fixed. - The latest production package no longer contains
serialport.node, and the old Event ID 1000 /0xc06d007ffailures stopped. However, the production app still cannot be used: it now initializes successfully and then exits silently after approximately 6–10 seconds. - Beta
26.715.3651.0continues to work, while production26.715.4045.0consistently exits on the same machine.
Therefore, the latest package appears to have fixed the original openai/codex#33375 failure mechanism, but it has not restored production-app usability on this system. The current silent-exit behavior is a distinct post-update failure rather than another observed serialport.node exception.
This is consistently reproducible with:
- Store package
OpenAI.Codex_26.715.4045.0_x64__2p2nqsd0c76g0 - Internal desktop release
26.715.31925(buildFlavor=prod) - Windows 11 Pro, build 26200, x64
The app gets well past early Electron startup:
- the packaged app-server is spawned;
- the initialization handshake succeeds;
- the main frame finishes loading;
- the window reaches
ready-to-show; - React routes mount and logs report
rendererWindowVisible=true; - the in-app browser backend reports ready;
- the log then ends abruptly after otherwise successful requests.
Windows AppModel Runtime shows the package container being created at 15:00:25 and destroyed at 15:00:31. There is no corresponding Application Error event, WER event, Defender/Code Integrity/AppLocker block, or Crashpad/LocalDumps dump.
The separately installed Beta package (OpenAI.CodexBeta 26.715.3651.0) works normally on the same machine.
This does not appear to be the previously reported serialport.node / 0xc06d007f issue in openai/codex#33375:
- the current production package contains no
serialport.nodefile; - no new Event ID 1000 /
0xc06d007fevent is generated; - the current failure is a silent full-process exit after the renderer is visible.
What steps can reproduce the bug?
- Install or update the production app from Microsoft Store to
OpenAI.Codex 26.715.4045.0on Windows 11 x64. - Fully terminate all production app processes. A Windows restart and launching production before Beta produces the same result.
- Launch the production app from Start, or activate it with:
``powershell``
Start-Process explorer.exe 'shell:AppsFolder\OpenAI.Codex_2p2nqsd0c76g0!App'
- Observe that the process tree starts and the window is initialized, but all production processes disappear after approximately 6–10 seconds.
- Reset only the production package and retry:
``powershell``
Get-AppxPackage -Name OpenAI.Codex | Reset-AppxPackage
- Confirm that the issue remains reproducible from a completely empty production AppX state. In my clean-reset test, the process count was 5 after 1 second, 8 after 3–5 seconds, and 0 after 10 seconds.
What is the expected behavior?
The production app should remain running and display an interactive main window after its app-server, renderer, and routes initialize successfully. It should not silently terminate the complete AppX process tree without an error or crash report.
Additional information
Troubleshooting already completed:
- Initially reproduced the older
26.707.9981.0production crash and used Codex Beta as the working fallback. - Updated to the latest production ChatGPT/Codex package after openai/codex#33375 was marked completed; production still exits, although the old
serialport.nodeexception is gone. - Windows Settings: Terminate → Repair → restart Windows → launch production first: no change.
- Backed up and then reset only
OpenAI.Codex; verified the package state was empty before relaunch: no change. - Package status after reset:
Ok. - Disabled GPU and GPU compositing: no change.
- Checked Windows Application, Defender, Code Integrity, AppLocker, AppModel Runtime, and crash dump locations.
- No Application Error Event ID 1000, WER crash event, security-policy block, or dump is generated for the clean-reset reproduction.
- The clean-reset log contains only two warnings, both saying an IPC broadcast had no configured handler; no fatal/error/shutdown record precedes termination.
- Production executable SHA-256:<br>
305B25FA057C35241C2C27BCB1112450F35EEE12C1D4B1E4D74C073454914346
Clean-reset timeline (UTC):
07:00:25.538 Launching app (prod, packaged, win32)
07:00:26.384 packaged app-server spawned
07:00:26.715 initialize handshake outcome=success
07:00:26.870 main frame finished load
07:00:26.896 window ready-to-show
07:00:30.601 app routes mounted; rendererWindowVisible=true
07:00:30.790 in-app browser backend startup ready
07:00:31.101 last log line: successful account/read response
07:00:31.xxx AppX container destroyed
Potentially related silent-exit report: openai/codex#22633. However, this reproduction reaches a visible, initialized renderer and generates normal desktop logs before the process tree disappears.
A sanitized diagnostic excerpt is attached. Raw logs are not attached because they may contain local paths, request IDs, and session metadata.
4 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I reviewed both issues suggested by the duplicate-detection bot.
26.707.9981.0and reports runtime download/connectivity/certificate failures. My reproduction uses production26.715.4045.0after a complete AppX reset; the app-server handshake, main frame, renderer routes, and browser backend all initialize successfully before the process tree exits.26.715.3651.0, and oneChatGPT.exeterminating with0xC0000005(Access Violation). I performed an additional exit-code capture on my Windows 11 build 26200 / production26.715.4045.0reproduction: all eight tracked production processes terminated with0x00000000, and no0xC0000005exit was observed.For those reasons, I believe this issue should remain separate unless maintainers confirm that the clean status-0 exit in
26.715.4045.0shares the same root cause as #33119.补充一个后续正式版仍可复现的环境:
OpenAI.Codex 26.715.8383.0 x6426.715.61943(prod)0.145.0-alpha.2710.0.26200.0,x64现象与本问题一致:登录可以完成,应用能够成功启动 app-server、完成初始化握手、加载主窗口并达到
ready-to-show,React 路由、账户读取和线程读取也能成功;随后整个 AppX 进程树在没有错误对话框的情况下静默退出。2026-07-22 多次复现,通常在启动后约 20–75 秒退出。Windows 事件日志只记录 AppX 容器创建和销毁,没有新的 Application Error 1000、WER 崩溃、Application Hang 或 Crashpad 转储。桌面日志在一系列成功请求之后突然结束,没有 fatal、崩溃或正常关机记录。日志中出现过 Statsig bootstrap 超时,但应用仍继续完成路由和账户初始化,因此它不一定是直接原因。
已经尝试:
elevated与unelevatedWindows 沙箱之间切换;均未解决。此前回退一个 Windows 更新后曾短暂恢复登录,但 Microsoft Store 将正式版从
26.715.4045.0更新到26.715.8383.0后,同类静默退出再次出现。当前没有证据表明这是用户项目或 Codex 配置导致的。请问该静默退出回归是否已经被官方跟踪?是否有计划在后续 Windows 正式版中修复?
It was discovered that the conflict was caused by a program named "SecureLink". I uninstalled it and now ChatGPT is functioning properly.