Claude Code To Codex

Claude Code → Codex Translation Mapping

A comprehensive, field-level mapping between [Claude Code](https://docs.anthropic.com/en/docs/claude-code) and [OpenAI Codex](https://openai.com/index/introducing-the-codex-app/) for practitioners migrating AI-agent operations stacks between the two systems.

**This is not a feature comparison.** It's a working reference that documents how every major system-level concept in Claude Code maps to its Codex equivalent — including exact config field names, CLI flags, file paths, and behavioral gotchas that only surface in practice.

**Last verified:** 2026-03-12 **Covers:** Claude Code (Sonnet 4.x / Opus 4.x era) and Codex CLI (v0.110.0–v0.114.0+)

Who this is for

You've built something on Claude Code — skills, automations, MCP integrations, subagent dispatch patterns — and now you need to transpose that system onto a client or team running OpenAI's stack. Or vice versa. This document makes that transposition efficient instead of painful.

The core insight

Both systems share the same fundamental paradigm: a single-agent event loop with skills-based extensibility, MCP for external tool integration, and local-first execution with optional cloud delegation. They diverge on runtime assumptions — Claude Code uses JSON config and `launchd` + `claude -p` for automation; Codex uses TOML config and a desktop app with built-in automations. The permission models, sandbox architectures, and context management strategies differ enough to produce silent failures if mapped naively.

Document structure

Sections (concept mapping)

#	Section	What it covers
1	Executive Summary	Paradigm overview, key behavioral differences
2	Configuration & Project Setup	CLAUDE.md → AGENTS.md, settings, MCP config, permissions, memory
3	Skills	SKILL.md format, discovery, activation, worked exampl