Claude Opus 4.6 (x3) always delegates some tasks as code review to cheapest models (x1)

### Describe the bug

I don't understand the reason for using Claude Opus 4.6, which costs 3 premium requests for some tasks, as `/review`, if it always delegates the whole work to Claude Sonnet 4.5, which costs 1 premium request.

So I'm always paying 3 requests for a poorer model like Claude Sonnet 4.5 in such cases.

### Affected version

1.0.20

### Steps to reproduce the behavior

1. Use `/review` when model Claude Opus 4.6 (high) is selected.
2. At the end, check the `/session info`
3. The whole work was done by the x1 model.

### Expected behavior

I'm expecting the selected model to be working on the review, because I'm expecting better results from it.

### Additional context

_No response_

Reproduce the behavior by selecting Claude Opus 4.6 (high), running /review, and checking /session info. Trace the /review command and model-delegation path to determine why the selected model is replaced; done when the selected model performs the review and session information reflects that model.

索引モデルが issue の本文から書いたものです。