claude-haiku-4.5 sub-agent fails with reasoning effort 'medium' not supported

Bug Report

Description

When the CLI internally routes sub-agent tasks to claude-haiku-4.5, it applies a medium reasoning effort, which is not supported by that model. This causes the following error:

Execution failed: Error: Reasoning effort 'medium' is not supported for model 'claude-haiku-4.5'

Steps to Reproduce

The error occurs during normal CLI usage when the task/explore sub-agent system selects claude-haiku-4.5 as the model with medium reasoning effort applied.

Expected Behavior

The CLI should not assign a reasoning effort level that is incompatible with the selected model. Either:

• Do not set reasoning effort for models that don't support it (e.g., claude-haiku-4.5)
• Map medium to a compatible default for claude-haiku-4.5
• Validate model/reasoning-effort compatibility before dispatch

Environment

• CLI Version: 1.0.79
• Model (main session): gpt-5.4
• Failing sub-agent model: claude-haiku-4.5
• OS: Linux

从报告中描述的 CLI 的 task/explore 子代理路由和模型选择路径开始，其中 claude-haiku-4.5 接收 reasoning effort。检查模型与 effort 的兼容性是如何处理的；完成的标准是，正常使用 CLI 时不再为该模型分发不受支持的 medium 设置；如果找到相关测试位置，还应加入回归检查。