是从 openai/codex 的代码里找出来的一些好玩的配置。 
寻找方法是把仓库克隆下来，然后让 Codex 它自己找有哪些配置能改的。 
~/.codex/config.toml： 
model = "gpt-5-codex"
model_reasoning_effort = "high"
model_reasoning_summary = "detailed"  # auto(默认) | concise | detailed | none，没感觉出有多细节但是我喜欢开着
model_verbosity = "high"              # low | medium(默认) | high，输出篇幅详细还是简略
sandbox_mode = "workspace-write"      # read-only(默认) | workspace-write | danger-fu...