Skip to main content

​
How configuration works

The plugin uses your local Codex CLI, which reads configuration from standard Codex config files. There are two levels:
  • User-level: ~/.codex/config.toml — applies to all repositories on this machine
  • Project-level: .codex/config.toml at the root of the directory you started Claude in — applies only to that project
Project-level config only loads when the project is trusted.
If you already use Codex directly, the plugin uses the same config you already have — nothing new to set up.

​
Setting the default model

Set the model key in your config file to control which model all plugin commands use by default: 
model = "gpt-5.4-mini"
This model is used by all plugin commands (/codex:review, /codex:adversarial-review, /codex:rescue) unless overridden per-command.

​
Setting reasoning effort

Set model_reasoning_effort to control how much reasoning the model applies: 
model_reasoning_effort = "xhigh"
Valid values are: none, minimal, low, medium, high, xhigh.

​
Example project-level config

Place this file at .codex/config.toml in the root of your project: 
model = "gpt-5.4-mini"
model_reasoning_effort = "xhigh"

​
Config file locations

LocationScope
~/.codex/config.tomlAll repositories on this machine
.codex/config.toml (project root)The current project only (must be trusted)

​
Per-command overrides

/codex:rescue supports --model and --effort flags to override config for a single run: 
/codex:rescue --model gpt-5.4-mini --effort medium investigate the flaky test
The spark model alias maps to gpt-5.3-codex-spark: 
/codex:rescue --model spark fix the issue quickly

​
Custom API base URL

If you need to point the OpenAI provider at a different endpoint, set openai_base_url in your Codex config: 
openai_base_url = "https://your-custom-endpoint.example.com/v1"
This is useful if you use an API key tied to a custom base URL. Because the plugin delegates through your local Codex CLI, your existing sign-in method and authentication state still apply — no additional setup is needed.
The full Codex configuration reference is available in the Codex docs.