Plan/Default mode toggles add repeated noisy model-change messages to the transcript
Summary
Switching session modes with Shift+Tab (or /plan) appends a visible model-change message to the conversation every time. Repeated toggling produces a large block of noise in the transcript.
Observed behavior
Each transition renders an entry such as:
- Model changed to gpt-5.6-terra medium for Plan mode.
- Model changed to gpt-5.6-terra default for Default mode.
After toggling back and forth, these entries accumulate rather than being collapsed, replaced, or kept out of the visible conversation.
<img width="988" height="908" alt="Image" src="https://github.com/user-attachments/assets/d045c274-ad72-48d7-be96-025073a78eea" />
Expected behavior
Mode changes should be unobtrusive. For example, show the active mode in the UI without adding a transcript item, or collapse or replace prior mode-change events so only the current state is visible.
Why this matters
Plan and Default are useful modes to switch between while working, but the growing sequence of status messages makes the conversation harder to scan and creates a noticeably poor UX.
Reproduction
- Start a Codex session.
- Press Shift+Tab to enter Plan mode.
- Press Shift+Tab again to return to Default mode.
- Repeat a few times.
- Observe a separate model-change message for every toggle in the transcript.
Thanks!