CLI: network drop during Thinking/Working shows no immediate disconnect indicator
Open 💬 3 comments Opened Feb 23, 2026 by ClarkOhlenbusch
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
What version of Codex CLI is running?
codex-cli 0.104.0
What subscription do you have?
Team
Which model were you using?
codex 5.3 xhigh (seen while the agent is in Thinking/Working)
What platform is your computer?
Darwin 25.1.0 arm64 arm
What terminal emulator and version are you using (if applicable)?
Apple Terminal (TERM_PROGRAM=Apple_Terminal, TERM=xterm-256color)
What issue are you seeing?
If internet connectivity drops while an active turn is still Thinking/Working, there is no immediate on-screen indication that the connection was lost. The UI keeps showing Thinking/Working as if the turn were progressing normally. This continues indefinitely.
What steps can reproduce the bug?
- Start Codex CLI and submit a prompt that keeps the turn active for several seconds.
- While the status row shows
Thinking/Working, disable internet connectivity (for example, turn Wi-Fi off). - Wait a bit (optionally re-enable connectivity).
- Observe that the status can continue to show
Thinking/Workingwithout an immediate disconnect/reconnect indication.
What is the expected behavior?
When stream/transport connectivity is lost mid-turn, the CLI should promptly show an explicit status (for example Reconnecting... or Connection lost) instead of only showing Thinking/Working.
Additional information
- This is different from resume-specific stale
Workingbugs; this report is about a live active turn when connectivity drops. - Possibly related issues:
- #12382 (resume can get stuck on
Workingafter unclean disconnect) - #12548 (reconnecting loop until timeout)
- #11680 (stream disconnect while working with retries)
- User impact: it looks like normal model progress even when transport has already dropped.
3 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Also experienced this multiple times after network disconnection. codex-cli 0.104.0 in linux
@ClarkOhlenbusch this is the fake-progress failure operators hate — transport dies mid-turn, the CLI keeps showing
ThinkingorWorkingas if the run were alive, and there is zero indication the connection dropped. The operator assumes work is progressing while the session is already dead.That is the worst class of runtime lie because it defeats every health-check pattern an operator could build on top.
Are you more blocked by detecting transport loss immediately so the run can be recovered, or by controls that pause the session and surface what actually happened before the operator wastes time waiting on a dead turn?