Codex mobile remote visibly reconnects on app resume instead of silently restoring active thread
What version of Codex CLI is running?
codex-cli 0.113.0
What Codex app package is installed?
OpenAI.Codex_26.602.4764.0_x64__2p2nqsd0c76g0What platform is your computer?
Microsoft Windows 11 Pro 10.0.26200 build 26200, 64-Bit
What surface/component is affected?
Codex Desktop App host + ChatGPT mobile Codex remote access
Which model were you using?
GPT-5 / Codex desktop thread
What mobile client were you using, if applicable?
ChatGPT mobile app using Codex remote access; exact mobile app version not provided
What issue are you seeing?
When the ChatGPT mobile app is backgrounded or loses focus and then returns to an active Codex remote thread, the client visibly reconnects instead of silently and reliably resuming the existing active thread. When this happens during an active assistant response, it appears related to duplicate rendering of the final assistant answer as reported in #26682. Even when no response is streaming, the repeated visible reconnect creates avoidable friction and makes the remote session feel unstable.
What steps can reproduce the bug?
Not fully deterministic yet, but repeatedly observed by a user controlling a Windows Codex Desktop App thread from ChatGPT mobile: 1. Open an active Codex remote thread in ChatGPT mobile. 2. Switch away from the ChatGPT app or let it lose focus. 3. Return to ChatGPT mobile. 4. The remote view reconnects visibly before the thread is usable again. 5. If the app switch happens while Codex is still streaming/responding, the same reconnect/resync path may render the final answer twice on mobile while the host transcript contains only one final answer (#26682).
What is the expected behavior?
Returning to the ChatGPT mobile app should silently and reliably resume the active Codex remote thread whenever the host session is still available. A technical reconnect may happen in the background, but the user should not have to experience a repeated visible reconnect for normal app switching, and reconnect/resync must be idempotent: no duplicate rendered assistant messages, no lost active thread state, and no manual recovery for quick focus changes.
Additional information
- Observed at: 2026-06-06T23:23:27.6003658+02:00
- Session/thread id: Not provided.
- Related issue: #26682
This helper intentionally does not attach logs, transcripts, secrets, repository contents, or screenshots. Add a screenshot manually only after checking it does not reveal sensitive data.
4 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I reviewed the suggested potential duplicates. This issue is related to #26682, #25606, and #26191, but I do not think it is a direct duplicate.
The specific behavior here is visible reconnect/resume friction whenever returning to an active Codex mobile remote thread after app focus/background changes, even when no stream is currently stale. During active responses, the same resume path may lead to the duplicate final-answer rendering tracked in #26682.
So this issue is meant to track the reconnect/resume UX and idempotency problem itself, not only one downstream symptom.
!image
Adding another repro from a macOS host + ChatGPT mobile remote session.
Environment:
ChatGPT.app/com.openai.codex26.715.31925(build5551)codex-cli 0.145.0-alpha.18Darwin 25.5.0 arm64 arm(Apple Silicon Mac)Observed behavior:
Expected behavior:
If the host remains available, returning to ChatGPT mobile should silently restore the active Codex remote thread. Any reconnect/resync should happen in the background and preserve active task state/notifications.
Host-side checks: