Compaction progress state not clearing in Windows app when compaction task fails (stuck "자동으로 컨텍스트를 압축하는 중")

Open 💬 3 comments Opened Mar 6, 2026 by MayoneJY

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

26.304.1143.0

What subscription do you have?

pro

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

During a turn that triggers compaction, the UI shows the in-progress message:
"자동으로 컨텍스트를 압축하는 중"
After the operation returns (including error paths), the message sometimes remains on screen and does not clear.
This appears like the compaction item is shown as started but never completed in the UI state.

I investigated and found that compaction lifecycle events may not always emit completion in all error paths, so the Windows app likely keeps the progress indicator in an active state.

What steps can reproduce the bug?

  1. Start a session where compaction is likely to run (auto compact threshold or manual /compact).
  2. Trigger a compaction path that can error
  3. Observe the Windows UI text/progress state:
  • "자동으로 컨텍스트를 압축하는 중" appears
  • Error/turn completion occurs, but the message is not dismissed.
  1. (If available) include:
  • 019cc08a-2afd-78e3-bbc0-d2e172225a07
  • 121k / 122k context token

What is the expected behavior?

The compaction progress should always complete in the UI by pairing start/completion events correctly.
Even on failure, the app should:

  • emit/handle compaction completion state
  • clear "자동으로 컨텍스트를 압축하는 중"
  • show an error message instead of a stuck-in-progress indicator

Additional information

I’m sharing this as a bug report plus root-cause analysis suggestion (no unsolicited code change, respecting the “invitation only” external contribution policy).

From investigation, this appears related to compaction lifecycle events: ContextCompaction start is emitted, but completion may not always be emitted on all failure paths, leaving clients that rely on started/completed pairing in a stale state.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗