Skip to content

Codex Mobile remote prompt stuck in Status Sending after steer blocks further input #36831

Open

Description

@harryshawk

Summary

A Codex Mobile remote-control thread can get into a state where a user prompt has apparently been accepted as a steer, but the mobile UI still shows it as a pending queued/sending item. The item cannot be opened or deleted, and the thread stops accepting new remote input from mobile.

This is user-input/control-surface loss rather than only a cosmetic status problem: the mobile client shows a stuck Status Sending... pill at the bottom of the thread, while the prompt composer is effectively unusable for new messages in that same thread.

Observed

Observed on 2026-08-04 from the ChatGPT mobile Codex remote-control surface connected to a host shown in the UI as mac-mini.lan.

  1. A Codex thread was active on the remote host.
  2. A prompt was entered from mobile while the thread was active.
  3. The prompt was steered rather than sent as a normal next turn.
  4. After that, the bottom of the mobile thread showed Status Sending... for that prompt.
  5. The prompt remained stuck in that state.
  6. The pending/stuck item could not be opened for inspection or editing.
  7. The pending/stuck item could not be deleted/canceled.
  8. New messages could not be entered into the thread from remote/mobile while this stuck state remained.

A screenshot of the failing state shows the mobile thread with a Status Sending... pill near the composer and the remote host context mac-mini.lan visible in the header.

Expected behavior

One of these should happen reliably:

  • If the mobile prompt is accepted as a steer, it should leave the sending/queued UI state and become visible as a delivered steer or reflected in the active turn.
  • If the steer cannot be delivered, the UI should show a failed/retryable state.
  • If the message is still pending, it should be inspectable and cancelable.
  • The mobile composer should not be permanently blocked by a stale pending item.
  • Queue/steer state should reconcile across mobile reload/reconnect and the remote execution host.

Actual behavior

The prompt appears stuck in Status Sending... indefinitely. It is not inspectable, editable, deletable, or recoverable from the mobile UI, and no new remote messages can be input into that thread from mobile.

Why this matters

Mobile remote control is often used to steer or queue work while away from the host machine. If a pending steer can become an undeletable, unopenable blocker, the user loses the ability to control or recover that active thread remotely. It also creates ambiguity about whether the host received the instruction, whether the instruction is still pending, or whether the thread is simply wedged in the client UI.

Related but not exact duplicates

Suggested fix direction

  • Treat mobile steer/queue submissions as durable records with stable client IDs and explicit states: local pending, server accepted, delivered, failed, canceled.
  • Give every pending mobile queue/steer item an inspect/cancel affordance.
  • On reconnect/thread hydration, reconcile pending mobile items against host/app-server state instead of leaving them in an indefinite sending state.
  • Ensure a stale pending item cannot block the composer forever; expose retry/cancel or automatically clear it after a known terminal failure.

Activity

  1. harryshawk commented on Aug 4, 2026

    @harryshawk
    Author

    Additional diagnostic context:

    • Affected Codex thread ID: 019fbf2d-6cdf-74d3-9ba9-b8649ce6931c

    This is the thread shown in the mobile remote-control surface where the prompt remained stuck in Status Sending... after being steered.

  2. added
    bugSomething isn't working
    iOSIssues related to the Codex iOS app
    sessionIssues involving session (thread) management, resuming, forking, naming, archiving
    connectivityIssues involving networking or endpoint connectivity problems (disconnections)
    on Aug 4, 2026
  3. ESKehler commented on Aug 30, 2026

    @ESKehler

    Update: the complete synchronized Android + Windows evidence is now consolidated in #41708.

    #41708 covers both queue-reconciliation outcomes captured on Android Remote:

    1. Force-stop recovery / queue loss: the Windows host had already recorded task_complete, but Android retained a stale Stop state and six pending messages. Force-stopping/reopening ChatGPT restored the UI but discarded the entire six-message queue.
    2. Delayed dequeue after task switching: two messages remained queued for approximately 38–41 minutes, then reached the host after switching to another task and returning.

    This issue remains related: both reports show mobile pending/steer state failing to reconcile with the authoritative host turn, leaving user-authored input stuck, lost, or delivered much later than represented.

    Related: #41708

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't workingconnectivityIssues involving networking or endpoint connectivity problems (disconnections)iOSIssues related to the Codex iOS appremotesessionIssues involving session (thread) management, resuming, forking, naming, archiving

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions