TL;DR

The issue might be related to the timing of the summary and the appearance of the "working" message, and adjusting the queue or message handling could potentially resolve it.

Guidance

• Investigate how the working message is being handled and displayed, and whether there's a delay or condition that's causing it to disappear when the model is summarizing.
• Review the code or configuration related to the agent's job and the queueing of messages to see if there's a way to prioritize or adjust the timing of the working message.
• Consider adding logging or debugging statements to understand the sequence of events and the state of the system when the working message is expected to appear.
• Look into the specifics of the gpt-5.4 model and its interaction with the Codex CLI to see if there are any known issues or quirks that could be contributing to this behavior.

Notes

The issue seems to be related to the interaction between the agent, the model, and the message queue, but without more information about the code or configuration, it's difficult to provide a more specific solution.

Recommendation

Apply workaround: Adjust the message handling or queueing to prioritize the display of the working message, as it's likely that the issue is related to the timing or sequencing of events rather than a fundamental problem with the Codex CLI or the model.