model_reasoning_effort = "high"
approval_policy = "on-request"
sandbox_mode = "workspace-write"
web_search = "live"