[Bug] Codex GPT models only show four reasoning levels and omit max

**提交人**: 匿名
**客户端版本**: 0.1.45

---

## Phenomenon

The Codex model configuration UI only offers four reasoning levels for GPT-series models:

`low`, `medium`, `high`, and `xhigh`.

The model catalog already declares `max` for GPT-5.6 models, but `max` is not available in the Codex configuration UI.

## Reproduction steps

1. Open Cindy's model configuration.
2. Select or configure a Codex GPT-series model.
3. Open the reasoning/thinking level selector.
4. Observe the available levels.

## Expected behavior

The UI should read the selected model's supported effort list (including any Codex-specific override) and show:

`low`, `medium`, `high`, `xhigh`, and `max` when supported.

## Actual behavior

Only four levels are shown; `max` is missing. Refreshing or restarting Cindy does not change the list.

## Frequency

Always reproducible.

## Tried

- Refreshed the model configuration.
- Restarted Cindy and checked again.

The current implementation appears to use a fixed Codex effort list instead of the model's declared `efforts` / `perAgent.codex.efforts`, so newer levels such as `max` are filtered out.
---
**OS**: darwin arm64 (25.5.0)
**界面语言**: zh-CN

Start by locating the Codex reasoning/thinking level selector and the model catalog effort resolution, including the perAgent.codex.efforts override mentioned in the report. Confirm the selector uses the selected model’s declared supported efforts, and consider the issue done when it displays max alongside the other levels whenever max is supported.

Written by the indexing model from the issue text.