TUI renderer switch (`/tui fullscreen`) silently reverts current-session model

Description

Running /tui fullscreen mid-session silently reverts the current session's runtime model to a previously used model, even though the persisted default in ~/.claude/settings.json still points at the newer choice. The user gets no warning and no confirmation prompt — subsequent responses come from the old model until they explicitly re-check via /model.

Environment

• Claude Code: 2.1.212
• macOS: 26.5.2 (build 25F84)
• Terminal: VS Code / Cursor integrated (TERM=xterm-256color, TERM_PROGRAM=vscode 3.12.17)

Steps to reproduce

1. Start a session where the persisted default is Opus 4.7.
2. Run /model and pick Opus 4.8 (1M context). Confirmation prints Set model to Opus 4.8 (1M context) and ~/.claude/settings.json now contains \"model\": \"opus[1m]\".
3. Run /tui fullscreen (renderer switch).
4. Run /model again — it reports Kept model as Opus 4.7.

Expected

Current-session model stays at whatever was selected before the TUI switch (Opus 4.8, matching settings.json).

Actual

Runtime model reverts to Opus 4.7 for the current session while settings.json still holds \"model\": \"opus[1m]\". New sessions still start on Opus 4.8 (persistence is intact); only the live session is affected.

Impact

Silent model downgrade. No visible cue that the model changed. Any responses in the same session after /tui fullscreen come from a different (older, less capable) model than the user believes they're using.

Workaround

Re-run /model and re-select the intended model after any /tui toggle. Or start a fresh session.

Note

claude --help in 2.1.212 doesn't expose a --tui / --renderer / --fullscreen CLI flag, so this issue is scoped to the interactive slash-command path.