子模型的思考模式无效

What version of Kimi Code is running?

0.38.0

Which open platform/subscription were you using?

OS: Windows

Which model were you using?

opencode/deepseek

What platform is your computer?

x64

What issue are you seeing?

Kimi Code CLI version: 0.38.0
OS: Windows

Problem:
A global [thinking].effort is automatically written into config.toml
when switching the main model through /model. This globally overrides
default_effort configured for aliases in [secondary_model.models].

What steps can reproduce the bug?

Steps to reproduce:

1. Define two aliases for the same model:
   • one with default_effort = "low"
   • one with default_effort = "high"
2. Add both aliases to [secondary_model.models].
3. Configure:
   [thinking]
   enabled = true
   and omit the effort field.
4. Start or reload the TUI.
5. Switch the main model through /model without explicitly changing
   the thinking effort.
6. Check config.toml.

Actual:
[thinking].effort = "xhigh" is written back automatically.
Subagents then use the global xhigh value instead of their alias-level
default_effort.

What is the expected behavior?

Expected:
Switching the main model should not create a global thinking.effort
unless the user explicitly changes the global default. Per-model
default_effort variants in the secondary model pool should continue
to work.

Additional information

No response

Contribution

• I am willing to submit a PR for this bug fix myself (please wait for maintainer approval in this issue first)

Start by reproducing the issue with config.toml, [thinking], [secondary_model.models], and the /model command described above. Trace the configuration update performed when switching the main model, then verify that an omitted global effort remains omitted and that aliases with low and high default_effort retain their respective behavior.