Desktop app hides task interruption/stall, making an inactive Codex run appear to still be working

Summary

Since updating to the new Windows desktop app experience, long-running Codex tasks no longer expose enough execution progress to distinguish active work from an interrupted or stalled run.

This has happened multiple times: the UI looked as if Codex was still working, but the underlying run had already stopped or become stuck. Because there was no clear terminal state or useful progress signal, I waited unnecessarily and could not steer the task at the point where intervention was needed.

The affected runs occurred in other task windows. This issue is being filed from a separate task, so the current conversation/session should not be treated as the failing sample.

Steps to reproduce

1. Start a long-running, tool-heavy Codex task in the Windows desktop app.
2. Let it run for an extended period.
3. Encounter an interruption, agent stop, tool failure, or stalled execution.
4. Observe that the UI continues to look like a task is working, or provides too little information to distinguish the stopped/stalled state from active reasoning.

Actual behavior

• Useful thinking/progress summaries are hidden or absent.
• An interrupted or stalled run may remain visually indistinguishable from an active run.
• There is no sufficiently prominent "stopped", "failed", "blocked", "needs input", or last-activity indication.
• Mid-run steering becomes impractical because the user cannot tell what stage the task reached or whether new instructions can still affect the active run.

Expected behavior

The desktop app should expose an observable execution state independently of private chain-of-thought, for example:

• active reasoning vs. active tool execution vs. waiting;
• last successful activity and elapsed time since that activity;
• explicit stopped/interrupted/failed/blocked/needs-input states;
• concise milestone or tool-status summaries;
• a clear indication of whether a follow-up message will steer the current run or be queued for a later run.

This request is not for raw chain-of-thought. It is for reliable operational visibility so users can distinguish genuine progress from a dead or interrupted run and steer Codex effectively.

Environment

• Surface: ChatGPT/Codex desktop app
• OS: Windows
• Timing: observed repeatedly after the recent desktop app update/merge
• Exact app version and affected session IDs: not available in this reporting task