Voice screen context cannot inspect the active Codex thread in the unified ChatGPT app

Summary

In the unified ChatGPT/Codex macOS app, a Voice-in-Codex session could identify the active Codex thread but could not access the visible message contents of that same thread.

Expected behavior

With Screen & System Audio Recording already enabled for ChatGPT, voice coordination should be able to capture sufficient app context to read the active Codex thread, or the product should clearly expose the required Appshot/computer-use action.

Actual behavior

The available screen-context path returned only thread metadata. A direct capture attempt was rejected by the backend execution environment, so the voice session could not see the message pane.

Why this matters

This makes it difficult to navigate a Codex task together by voice: the user can see and dictate into the active task, while the assistant cannot verify the on-screen content.

Requested outcome

Please support explicit, reliable screen/Appshot capture for the active Codex thread from Voice in the unified ChatGPT app, or provide clear in-product guidance and diagnostics when that context is unavailable.

Environment

• macOS unified ChatGPT/Codex desktop app
• Voice in Codex
• Screen & System Audio Recording enabled for ChatGPT

Start by reproducing the Voice in Codex screen-context path in the unified macOS ChatGPT/Codex app with Screen & System Audio Recording enabled, checking whether it returns only thread metadata. Trace the Appshot or computer-use capture entry point and backend rejection; done means the active thread's message pane is reliably available, or the app provides clear guidance and diagnostics when it is not.