[macOS][gpt-6-astra] Cannot attach requested images in question dialogs

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

26.901.31953 (build 7868)

What subscription do you have?

API billing

What platform is your computer?

Darwin 25.6.0 arm64 arm macOS 26.6.1 (build 25G76), Apple Silicon.

What issue are you seeing?

While using GPT-6 Astra in the Codex desktop app, the assistant frequently asks follow-up questions through the structured question dialog. Some of those questions ask me to provide an image or screenshot, but the question interface does not let me attach the requested image.

To provide it, I have to dismiss or deny the question first, return to the normal chat composer, and send the image as a separate chat message. I want to answer the question, but the available workflow requires me to reject or dismiss it before I can supply the requested input.

This creates a mismatch between what the assistant requests and what the question interface accepts. Frequent follow-up questions make the interruption more disruptive: each image-dependent question requires an extra transition back to normal chat.

Current environment, verified while preparing this report:

Field	Value
Model	GPT-6 Astra
Model identifier	gpt-6-astra
Reasoning effort	medium
Model provider	openai
Collaboration mode	default
Desktop app version	26.901.31953
Desktop app build	7868
Bundled Codex runtime	0.153.1
Application bundle identifier	com.openai.codex

The model identifier and reasoning effort were checked against the current session's recorded turn metadata, rather than inferred from the assistant's self-description. These are the reporting session's settings; an earlier failing question's exact tool payload has not been captured.

What steps can reproduce the bug?

1. Open the Codex desktop app and use GPT-6 Astra (gpt-6-astra). The current reasoning setting is medium.
2. Start or continue a task that requires an image or screenshot from the user.
3. When the assistant asks for that image through the structured question dialog, try to provide it as the answer.
4. Observe that the question interface offers no way to attach the requested image.
5. Dismiss or deny the question.
6. Return to the normal chat composer and attach the image there as a separate message.

These steps describe the observed workflow. The original triggering prompt has not been preserved in this report, and the model may not choose the question dialog on every attempt.

What is the expected behavior?

Users should be able to supply an image the assistant requests without dismissing or denying the question.

Either of the following would address the workflow:

• Support image attachments in the question response interface, retaining the image and text as the answer to that question.
• If that interface cannot accept images, route image requests through the normal chat composer or provide a direct way to answer there while keeping the question pending.

The assistant should use a response interface that supports the input it asks for. Supplying the requested image should complete the question normally.

Additional information

No response

Start by locating the structured question dialog and the normal chat composer in the macOS desktop app, then trace how response input types and attachments are handled. Done means a requested image can be supplied without dismissing the question, either within the dialog or through a flow that keeps the question pending; add or update coverage for that workflow.