codex
Codex: High-Reasoning AI Assistant for Claude Code
Use OpenAI's Codex CLI (gpt-5.5, xhigh reasoning) for complex coding, architecture, and review work that benefits from a frontier reasoning model. Always invoke via codex exec (Claude Code's bash environment is non-interactive).
Defaults (apply unless the user overrides)
| Parameter | Default | CLI flag |
|---|---|---|
| Model | gpt-5.5 |
-m gpt-5.5 |
| Sandbox | read-only |
-s read-only |
| Reasoning effort | xhigh |
-c model_reasoning_effort=xhigh |
Network access (with workspace-write) |
enabled | -c sandbox_workspace_write.network_access=true |
| Web search | built-in on supported models (no flag) | — |
Switch to workspace-write only when the user explicitly says "edit", "modify", "save", "write changes", "fix", "refactor", etc. When switching, always include the network_access=true config so package managers (npm/pip/cargo), git fetch, and HTTP calls work.
Use gpt-5.5-fast only when the user explicitly asks for speed ("fast", "quick"). On ChatGPT-account auth, only gpt-5.5 is callable; gpt-5.5-fast, gpt-5.5-codex, and gpt-5.5-pro require API-key auth.
Fallback chain: model gpt-5.5 → gpt-5.5-fast; effort xhigh → high → medium.