Thanks @lanternian for the PR, this looks promising.

Could you help add some tests before we merge? I’d like to cover the auto headers, and also the model-selection behavior when the existing Codex model is not in available_models.

We should make sure --auto actually uses an auto-available model instead of silently keeping or falling back to a non-auto model.

Thanks for your reply. Based on your advise I make some work:

I added a test auto-mode.test.ts to verify that when selecting a non-auto model under auto mode, it returns a 400 error instead of silently falling back to other models.

In additon, auto-mode.test.ts also verify these following feature:

When selecting Auto mode, the prompt's selectable models should only include auto-available models, preventing users from selecting non-auto-available models.

In Auto mode, the copilot-session-token is only carried when sending chat/response requests; in all other routers, it stays as the original default.

When a user uses --model to override the model, if the specified model is not in the auto-available list, a API Error: 400 Requested model not available for session error is returned, keeping the same behavior as before.