Voice overlay suppresses inline chat content

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.901.51231

What subscription do you have?

Enterprise

What platform is your computer?

Darwin 25.6.0 arm64 arm

What issue are you seeing?

During an active Codex Desktop voice conversation, the assistant sent several inline Markdown display items, including simple tables and plain text. The content did not appear in the chat timeline while the voice overlay remained active. Instead, the voice frontend spoke summaries as though the content had rendered.

The user reports that the missing content may appear only after ending or dismissing the voice interaction.

Actual behavior

• Assistant emitted inline display content multiple times.
• No table or plain-text content appeared in the active chat view.
• The user saw only the spoken voice interaction.
• The assistant had no reliable way to verify whether the display item reached the timeline.
• A visible-state check identified the active Codex task view but exposed no rendered timeline or queued-inline-content state.

What steps can reproduce the bug?

Minimal reproduction

1. Open a Codex Desktop task and start voice mode.
2. Ask the assistant to show a small Markdown table in chat and not read it aloud.
3. Have the assistant emit inline Markdown.
4. Observe that no inline table appears while the voice overlay is active.
5. End or dismiss the voice overlay.
6. Check whether the previously emitted content appears afterward.

What is the expected behavior?

Inline assistant Markdown should appear in the visible chat timeline immediately during an active voice session. If the UI intentionally defers rendering until voice ends, it should clearly indicate that content is queued and should not present it as already visible.

Additional information

Impact

This blocks tasks that require inspectable on-screen content during voice interaction: tables, links, code, plans, diffs, and written notes. It also causes misleading confirmations because the assistant can follow the inline-display protocol without receiving delivery confirmation.

Diagnostic conclusion

This appears to be a presentation or timeline-queue defect in the voice-overlay path, rather than a Markdown-authoring failure. The system accepted display-oriented output but did not expose a rendered-state acknowledgement or failure signal to the assistant.

Suggested telemetry to inspect

• Whether inline timeline events are accepted by the realtime backend.
• Whether the Desktop voice-overlay compositor queues or suppresses those events.
• Whether queued items are flushed when voice mode ends.
• Whether the frontend's spoken-response path incorrectly treats an inline item as delivered.
• Whether the client can report an explicit inline-render acknowledgement or failure.

Start by reproducing the issue in a Codex Desktop task with voice mode active, then trace the voice-overlay path and handling of inline timeline events. Check whether events are queued, suppressed, or flushed when the overlay ends; done means Markdown content appears immediately during voice mode or the UI clearly identifies it as queued.