# 1) Core model and runtime policy
# ------------------------------
model = "gpt-5.4"
review_model = "gpt-5.4"
model_provider = "custom" # or "openai"

approval_policy = "never" # untrusted | on-failure | on-request | never
sandbox_mode = "workspace-write" # read-only | workspace-write | danger-full-access
web_search = "cached" # disabled | cached | live
service_tier = "fast" # fast | flex
person...