How to use Claude Code and Codex together
To use Claude Code and Codex together, run each CLI in its own terminal on the same project, give each a clear role (for example, Claude Code writes and Codex reviews), and pass work between them. You can relay by hand, by files, or, in BRIX, by linking the two terminals so they send each other prompts with medmaestri ask and send.
Why use Claude Code and Codex on the same project?
Two different models catch different mistakes. An agent reviewing its own work tends to agree with itself; a second agent from another vendor, trained differently, reads the same diff with fresh eyes. Many developers already pay for both — a Claude plan and a ChatGPT plan — and use only one at a time.
Common ways to split the work:
| Pattern | Claude Code | Codex |
|---|---|---|
| Writer / reviewer | Implements the change | Reviews the diff and reports findings |
| Reviewer / writer | Reviews and challenges the plan | Implements the change |
| Split by area | Backend | Frontend (in a separate worktree) |
| Second opinion | Proposes a fix for a bug | Investigates the same bug independently, then the answers are compared |
Which model should write and which should review is a judgment call that changes with every model release. Try both orders on your codebase instead of trusting anyone's ranking.
Can Claude Code and Codex work in the same folder?
Yes, as long as only one of them is editing at a time. The safe default for the writer/reviewer pattern is one checkout: the writer edits, the reviewer only reads (git diff, the files, the tests). Codex has a built-in review flow for exactly this — /review runs against uncommitted changes, a commit or a base branch and reports prioritized findings without modifying your working tree (per the Codex CLI docs, checked October 2026).
If both agents will write code at the same time, give each its own git worktree so their edits only meet when you merge.
- Keep shared instructions in a file both agents read. Claude Code reads
CLAUDE.mdand Codex readsAGENTS.md; a common approach is to keep the rules inAGENTS.mdand reference it fromCLAUDE.md. - Decide who is allowed to commit. Two agents committing to the same branch is how history gets messy.
- Make the reviewer's output concrete: file, line, what breaks, how to reproduce.
How do I make one agent review the other without copy-paste?
There are three levels, from manual to connected:
- By hand. Two terminals side by side; you copy the summary from one into the other. Works, but you become the message bus and the bottleneck.
- By files. The writer leaves its plan in
PLAN.md, the reviewer writesREVIEW.md, and you tell each one to read the other's file. Better, but still you nudging each step. - By shelling out. One agent calls the other non-interactively, for example Claude Code running
codex exec "review the uncommitted changes". Useful for one-shot reviews; each call starts a fresh Codex session, so there is no running conversation to follow up in.
What none of these gives you is two live sessions that can talk to each other while you watch both. That is the gap BRIX fills.
How do Claude Code and Codex talk to each other in BRIX?
In BRIX, each agent is a real terminal running the CLI you already have — claude in one, codex in another — on your own Claude and ChatGPT logins. When you draw an edge between two terminals on the canvas, that edge is the permission: the two can now reach each other through the medmaestri command, which BRIX puts on the PATH of every terminal.
| Command | What it does |
|---|---|
medmaestri list | Lists the terminals linked to this one |
medmaestri ask "Codex" "..." | Sends a prompt to the Codex terminal and waits until it goes quiet, then returns its output |
medmaestri send "Codex" "..." | Types the text plus Enter into the other terminal and returns immediately |
medmaestri check "Codex" --clean | Reads what is on the other terminal's screen right now, without ANSI codes |
medmaestri spawn --cli codex --name "Codex" | Starts a new Codex terminal already linked to the current one |
Because the agents run these commands themselves, you can tell Claude Code "when you finish, ask Codex to review the diff and fix what it finds", and the round trip happens between the terminals. You see both sessions in full, and you can type into either one at any time.
How do I set up a Claude Code writer and a Codex reviewer?
- 1
Install BRIX and both CLIs
Download BRIX for macOS (Apple Silicon) or Windows. Have Claude Code and Codex installed and logged in; BRIX uses them as they are and doesn't charge for the models.
- 2
Open the project and start Claude Code
Open a terminal in your repository and run
claude. This one writes the code. - 3
Recruit Codex as the reviewer
From the Claude Code terminal (or ask Claude to do it):
medmaestri spawn --name "Codex" --cli codex --prompt "You are the reviewer. Do not edit files. When asked, review git diff and report findings with file and line.". The new terminal is born linked to Claude Code. Alternatively, open a Codex terminal yourself and draw the edge on the canvas. - 4
Give Claude Code the loop
Tell it: "Implement the change. Then run
medmaestri ask Codex "Review the uncommitted changes", fix what is valid, and tell me what you rejected and why." - 5
Watch, intervene, commit
Both terminals stay visible on the grid or canvas, and the Scene view shows which one is working or waiting for you. When the loop settles, review the diff yourself and commit.
Want the roles pre-drawn? The blindagem squad creates a security auditor, a diff reviewer and a cleanup agent already connected: medmaestri squad blindagem.
Does this work for more than two agents?
Yes. A lead agent can recruit up to five workers, one level deep, and each can run a different CLI — Claude Code, Codex, Antigravity or OpenCode. The canvas also holds notes and task lists the agents can read and update, so the plan lives in one place instead of in each chat. See Orchestrate a team of AI agents.
And the same pair can keep going after the merge: with BRIX PRO, the studios for copy, ads, carousels and video read the same local product base, so the launch material is written by agents that know what was just built. See the plans.
Frequently asked questions
Do I need both a Claude and a ChatGPT subscription?
Yes, to run both CLIs you need access to each one, through a subscription or API key. BRIX uses those logins as they are and does not charge for tokens.
Can Codex review code written by Claude Code?
Yes. Codex can read the same working tree, and its /review command reports findings on uncommitted changes, a commit or a branch without modifying files.
Will Claude Code and Codex overwrite each other's changes?
They can if both edit the same files at once. Keep one as a read-only reviewer, or give each its own git worktree when both need to write.
What is medmaestri?
It is the command-line tool BRIX installs in every terminal it opens. Agents use it to list linked terminals, send prompts with ask and send, read another terminal with check, and recruit new agents with spawn.
Can two terminals in BRIX talk without an edge between them?
No. A terminal can only reach terminals it is linked to; drawing the edge on the canvas, or recruiting with spawn, is what grants access.
Keep reading
- How to run multiple Claude Code agents in parallelHow to run multiple Claude Code agents in parallel: terminal tabs, tmux, git worktrees, Conductor, Claude Squad, and connected agents with a lead in BRIX.
- How to orchestrate a team of Claude Code agentsOrchestrate Claude Code and Codex agents as a team: a lead agent, canvas edges as permissions, ready-made squads, a task list and a CLI to spawn, ask and check.
- BRIX MCP: operate BRIX from Claude Desktop, Codex or Claude CodeOperate BRIX from Claude Desktop, the Codex app or Claude Code with the BRIX MCP server and the medmaestri CLI: spawn agents, send prompts, build squads.