Windows: local conversation archive fails with os error 2 after resume

Start at the thread/archive method and inspect how the referenced rollout JSONL path is handled on Windows after resume, including the local archived_sessions directory. Reproduce the archive failure and compare the normal and extended-length paths; done means an affected conversation moves once, disappears from the active list, and no longer reports os error 2.

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

Codex Desktop release 26.814.41407 (updated 2026-08-18; release observed in desktop log)

What subscription do you have?

ChatGPT Plus (currently using credits)

What platform is your computer?

Windows 10 Home 22H2, OS build 19045.7663, x64

What issue are you seeing?

Summary

After the 2026-08-18 update, Codex Desktop on Windows cannot archive local conversations. Older and current local project threads are visible and readable, but Archive fails and the thread remains active.

Error

failed to archive session: thread-store internal error:
failed to archive thread:
The system cannot find the file specified. (os error 2)

The desktop log reports method=thread/archive with error code -32603.

The referenced rollout JSONL file exists on disk, and the local archived_sessions directory also exists. No session files, database rows, or logs were modified during diagnosis.

What steps can reproduce the bug?

1. Start Codex Desktop on Windows.
2. Open or resume an existing local project conversation.
3. Use the conversation menu and choose Archive.
4. The app reports the archive failure shown above; the thread stays visible and unarchived.
5. Restart Codex Desktop and retry. The same error occurs for several older and current local threads.

What is the expected behavior?

The conversation should be marked as archived, its rollout file should be moved to the local archived-session store, and it should disappear from the active thread list.

Additional information

This appears to match open Windows reports:

• https://github.com/openai/codex/issues/39239
• https://github.com/openai/codex/issues/39209
• https://github.com/openai/codex/issues/39150
• https://github.com/openai/codex/issues/39130

Those reports describe a likely duplicate-path handling issue: one rollout file may be represented both as a normal Windows path and as an extended-length path after resume. The archive operation may queue the same physical file twice; the second move then fails with os error 2, and the first move is rolled back.

Could you confirm whether a fix is planned and provide a safe way to archive existing affected local conversations without manual database edits?