Windows Codex app crashes on startup when user profile path contains Korean characters #27707

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

Affected: 26.608.1337.0 Also tested: 26.609.3341.0, still crashes Last known working: 26.602

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?

[What issue are you seeing?]

Codex Windows app crashes on startup after updating to v26.608.1337.0.

The issue appears to happen when the Windows user profile path contains Korean characters.

Environment:

• OS: Windows 11 x64
• Codex version affected: 26.608.1337.0
• Last known working version: 26.602
• Windows user profile path contains Korean characters: C:\Users[Korean username]
• System ANSI code page: Korean CP949 environment

Symptoms:

• Codex crashes shortly after launch.

• The app log only shows a few early startup lines:

  • Launching
  • Windows Store update check
  • Hotkey registration

• Many empty .staging-* directories are repeatedly created under runtimes/cua_node/.

• Crashpad .dmp files are generated repeatedly.

• The app appears to attempt downloading or initializing this runtime:

cua-node-0.0.1-20260604060535-d6b2cf834a7a-pr-991379-win32-x64.zip

The crash appears related to a character encoding conversion failure:

__char_to_wide: Illegal byte sequence

What steps can reproduce the bug?

[What steps can reproduce the bug?]

1. Use Windows with a non-ASCII username, for example a Korean username.
2. Ensure the Windows user profile path contains Korean characters, such as C:\Users\[Korean username].
3. Install or update Codex to v26.608.1337.0.
4. Launch Codex.
5. Observe that Codex crashes shortly after startup.
6. Check that empty .staging-* directories are created under runtimes/cua_node/.
7. Check that Crashpad .dmp files are generated repeatedly.

What is the expected behavior?

[What is the expected behavior?]

Codex should launch normally even when the Windows user profile path contains non-ASCII characters such as Korean, Japanese, Chinese, or accented Latin characters.

[Additional information]

Starting from Codex v26.608, the newly added cua_node runtime code may be receiving a UTF-8 encoded path from Node.js and then passing it to native code that converts it using the system ANSI code page.

On Korean Windows, the system ANSI code page is CP949. If UTF-8 bytes for the Korean user profile path are interpreted as CP949/EUC-KR, the conversion can fail with an illegal byte sequence error.

This may cause the Codex app to crash during startup while initializing or downloading the cua_node runtime.

Possible workarounds:

• Use a Windows account with an ASCII-only profile path.
• Enable the Windows beta UTF-8 system locale option.
• Use Codex Web or Codex CLI instead of the Windows app.

This looks like a Unicode path handling bug in the Windows desktop app or the cua_node runtime initialization code. Native Windows APIs should ideally use wide-character paths directly instead of converting UTF-8 paths through the system ANSI code page.

Additional information

No response