Expose explicit Queue and Steer controls for Voice follow-ups to active Codex tasks

Summary

Voice coordination in the unified ChatGPT/Codex desktop app can send a follow-up prompt to an active Codex task, but the coordinator API does not expose whether that prompt should be queued after current work or should steer/interrupt the active task.

Expected behavior

Voice should support explicit, user-directed choices equivalent to:

• Queue: preserve the active task's current work and deliver the follow-up when the current step/turn completes.
• Steer: deliver the follow-up promptly and allow it to redirect the active work.

The selected delivery mode should be observable and confirmed to the user.

Actual behavior

The available follow-up message control accepts only a prompt and thread target. The runtime decides when and how to apply it, with no explicit queue/steer option exposed to the voice coordinator.

Why this matters

During voice coordination, users need to preserve a long-running task while safely lining up the next instruction. Without an explicit delivery mode, the assistant cannot reliably honor "queue this, don't interrupt" versus "steer it now."

Requested outcome

Expose queue and steer as explicit follow-up delivery modes for Voice in Codex, including a clear status/readback of which mode was selected.

Start by tracing the coordinator API's follow-up message control and the runtime path that applies prompts to active Codex tasks. Confirm how Voice selects and reports delivery today. Done means queue and steer are explicit modes, and the selected mode has clear user-visible status or readback.