Codex mobile remote renders one assistant final reply twice while local transcript has one final_answer
What version of Codex CLI is running?
Host Codex App: OpenAI.Codex_26.602.4764.0_x64__2p2nqsd0c76g0
CLI on PATH: codex-cli 0.113.0
What subscription do you have?
Plus
Which model were you using?
GPT-5.5
What platform is your computer?
Host: Windows 11 Pro 10.0.26200, x64.
Remote client: ChatGPT mobile app using Codex mobile/remote access. Exact mobile app version not available at report time.
What issue are you seeing?
A single assistant final reply appeared twice in the mobile Codex/ChatGPT remote view.
Important detail: the Windows host/local session transcript does not appear to contain the final reply twice. The local transcript for the affected turn shows:
- one
commentaryagent message with different short progress text - one
final_answeragent message with the actual answer - no second identical
final_answerin the local JSONL transcript
So this looks like a mobile remote rendering/sync duplication, not duplicate user input and not the known CLI/TUI case where identical content is emitted as both commentary and final_answer.
Possibly related but not the same root cause: #15633.
What steps can reproduce the bug?
Not deterministic yet. Observed once in an active Codex App thread controlled from the ChatGPT mobile app:
- Use ChatGPT mobile app to continue/control a Codex App thread on a Windows host.
- Ask a normal follow-up question.
- Codex sends a short progress/commentary message, uses local tools, then sends one final answer.
- On mobile, the final answer is rendered twice.
- On the host/local transcript, the final answer is present only once.
Affected local thread/session id: 019e94e6-ed3a-73b1-a314-1b702804301b.
Observed around: 2026-06-05 22:02 UTC / 2026-06-06 00:02 Europe/Berlin.
What is the expected behavior?
The mobile remote view should render exactly one visible copy of a final assistant reply when the local/session transcript contains only one final answer for that turn.
Additional information
I do not have a mobile screenshot attached yet. The report is based on the user's mobile observation plus a local transcript check on the Windows host immediately afterward.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗