OpenAI Codex + Claude Code Guide

OpenAI Codex and Anthropic’s Claude Code are both agentic coding tools. Each can read a codebase, edit files, run commands, and review changes.

Using both gives you:

• A second opinion on important changes
• Independent reviews from different model families
• A backup when one tool hits a usage limit
• The strengths of each tool’s ecosystem
• Parallel work on separate tasks

Two agents are not twice as good by default. They are better when each has a clear role.

Made by

OpenAI

Surfaces

ChatGPT desktop app, CLI, IDE extension, Codex Cloud, GitHub, Slack, mobile

Recommended models

GPT-6.1 Sol for complex work, GPT-6 Luna for focused tasks, GPT-6 Astra for the hardest problems

Instructions file

AGENTS.md

Skills folder

.agents/skills

Plans

Included from ChatGPT Plus ($20/month); Free and Go get GPT-6 Luna in the desktop app, subject to rollout

Made by

Anthropic

Surfaces

Terminal, VS Code, JetBrains, Claude desktop app, web, mobile, Slack

Models

Claude Opus 5.5, Sonnet, Haiku, and others, with effort levels from low to max

Instructions file

CLAUDE.md, and it can read AGENTS.md

Skills folder

.claude/skills

Plans

Included in all paid Claude plans, from Pro at $20/month

Windows installers, Homebrew, and npm options are available for both. Run codex and claude in your project folder to sign in.

Keep one source of truth so both agents follow the same rules.

• Write your project rules in AGENTS.md
• Codex reads AGENTS.md automatically
• Claude Code reads AGENTS.md when there is no CLAUDE.md (v2.1.277 or later)
• If you also need CLAUDE.md, import the shared file by adding @AGENTS.md to it

One rulebook, two agents. Never maintain two copies of the same rules.

Include in AGENTS.md:

• What the project does
• Setup and install commands
• Test, lint, and build commands
• Code style and naming rules
• Folders and files to avoid
• Actions that need human approval
• How to write commit messages
• What done looks like

Keep it short and specific. Codex caps combined AGENTS.md content at 32 KiB by default.

OpenAI publishes an official open-source plugin that runs Codex from inside Claude Code. It uses the Codex CLI already on your machine, with your existing login and settings.

Install it in Claude Code:

/plugin marketplace add openai/codex-plugin-cc /plugin install codex@openai-codex /reload-plugins /codex:setup

It requires a ChatGPT subscription (including Free) or an OpenAI API key, and Node.js 18.18 or later.

/codex:review

A standard read-only review of your current changes

/codex:adversarial-review

A review that challenges design decisions and assumptions

/codex:rescue

Hand a bug investigation, fix, or follow-up task to Codex

/codex:transfer

Turn the current Claude Code session into a Codex thread

/codex:status

See running and recent Codex jobs

/codex:result

Show the output of a finished job

/codex:cancel

Stop a running background job

The simplest high-value workflow: one agent builds, the other reviews.

• Claude Code implements the change
• Run /codex:review or /codex:adversarial-review
• Claude Code addresses valid findings
• You review the final diff and tests

Or reverse it: Codex builds, and Claude Code reviews the diff.

Treat each review as evidence to check, not a verdict to accept.

For larger features:

• Ask one agent for a detailed plan
• Ask the other to critique the plan
• Merge the strongest ideas yourself
• Let one agent build from the final plan
• Let the other review the result

Disagreement between agents is useful. It shows you where the real decisions are.

Never let two agents edit the same files at the same time.

• Give each agent a separate Git branch
• Use separate Git worktrees for parallel tasks
• Assign clearly separate tasks
• Merge one change at a time
• Run tests after every merge

Codex has /worktree to run a chat in a new Git worktree.

Codex Cloud

Run several well-scoped tasks remotely, then review the results.

Codex GitHub review

Mention @codex review on pull requests or turn on automatic reviews.

Claude Code skills and hooks

Package workflows and enforce checks before or after actions.

Claude Code routines

Run scheduled work in the cloud, even when your computer is off.

Both

MCP connections, subagents, and terminal workflows.

Test both on your own codebase. Real tasks reveal more than benchmarks.

Both tools use a SKILL.md file with a name and description.

• Codex reads skills from .agents/skills
• Claude Code reads skills from .claude/skills
• Keep the shared logic in scripts both can run
• Keep tool-specific notes short

Check each tool’s documentation for format differences before copying skills between them.

Both support the Model Context Protocol, so you can connect the same documentation, issue tracker, or database to each.

Codex

codex mcp add

Claude Code

claude mcp add

Give each agent only the servers its role needs.

Both agents should ask before pushing, deploying, deleting, or touching secrets.

• Use the cheapest model that does the job well
• Raise effort only for hard problems
• Use reviews on important changes, not every typo
• Switch tools when one hits a limit
• Keep instructions concise to save context

Codex uses /model to change models and reasoning effort. Claude Code uses /model and /effort.

Personally review:

• Architecture decisions
• Security-sensitive code
• Database migrations
• Payments and authentication
• Anything that deploys to production
• Disagreements between the agents

Both companies warn that their agents can make mistakes. Two agents agreeing is not proof.

Rules

One shared AGENTS.md.

Plan

One agent plans, the other critiques.

Build

One agent implements on its own branch.

Review

The other agent reviews the diff.

Fix

Address valid findings.

Test

Run the full test suite.

Approve

A human reviews and merges.

Learn

Update AGENTS.md with new lessons.

Give this to the building agent:

Read AGENTS.md and follow it exactly. Goal: [describe the change]. Constraints: [performance, security, or style requirements]. First, write a short plan and list the files you expect to change. Wait for my approval. Then implement the change on a new branch, add or update tests, and run the full test suite. When finished, summarize what changed, why, and what a reviewer should check most carefully. Do not push, deploy, or delete anything without asking me.

Then ask the other agent to review that summary and the diff, focusing on correctness, security, and missing tests.