Overview
OMX keeps the OpenAI Codex CLI as the execution engine and wraps a workflow around it. Instead of replacing the agent, it makes a session start stronger by default, gives you one consistent path from clarification to completion, and keeps project guidance, plans, logs and state under a `.omx/` directory in the repo. The canonical workflow commands — `$deep-interview`, `$ralplan`, `$ultragoal`, `$team`, `$code-review`, `$ultraqa` — are invoked independently, not as a fixed chain.
The recommended day-to-day flow is clarify, approve a plan, then run `$ultragoal` as the durable multi-goal completion wrapper, reaching for `$team` only when a story genuinely needs coordinated parallel work and `$ralph` when you want a single-owner completion loop instead. Research has its own bounded lanes: `$best-practice-research` for ordinary upstream evidence gathering and `$autoresearch` / `$autoresearch-goal` for validator-gated research artifacts, all of which feed back into `$ralplan` for architecture synthesis.
Isolation is taken seriously. `--worktree` launches into a separate git checkout under `../<repo>.omx-worktrees/`, which is the safer default when combined with `--madmax` (OMX shorthand for Codex's `--dangerously-bypass-approvals-and-sandbox`, so it removes approval and sandbox guardrails and belongs only in trusted repositories). Each standard launch owns one writable session pointer under its `OMX_ROOT`, so a second launch from the same checkout fails closed rather than silently sharing state. macOS and Linux are the actively tuned path; native Windows and the Codex App are explicitly not the default experience.
What it does
- Canonical workflow commands — `$deep-interview`, `$ralplan`, `$ultragoal`, `$team`, `$code-review`, `$ultraqa` — each usable independently
- Git-worktree launch isolation, named or detached, with fail-closed session-root ownership per checkout
- Agent teams over a durable tmux runtime (psmux on Windows), for coordinated parallel work inside one goal
- Durable state under `.omx/` for plans, logs, memory and mode tracking, plus scoped `AGENTS.md` project guidance
- Reasoning-effort shorthands (`--high`, `--xhigh`) and `--madmax` for Codex's bypass mode, with worktree launch recommended alongside it
- `omx doctor` install-shape checks plus an `omx exec` smoke test that proves the live Codex runtime can authenticate and complete a call
- An official Codex plugin layout at `plugins/oh-my-codex` with marketplace metadata, alongside the global CLI
Getting started
Requirements: Node.js 20+, an installed and authenticated Codex CLI (`codex --version`), and tmux on macOS/Linux for the recommended durable team runtime. Do not reinstall @openai/codex through npm if Homebrew already owns the codex binary — npm can fail with EEXIST.
Install OMX
If Codex CLI is already installed, just add OMX. Otherwise let npm manage both.
npm install -g oh-my-codex
# or, if you have no Codex CLI yet:
npm install -g @openai/codex
npm install -g oh-my-codexChoose a setup scope
Project scope makes the repository own the durable AGENTS.md guidance; user scope configures Codex at the user level. Avoid running project-scoped setup from your home directory.
omx setup --scope project --merge-agents
# or: omx setup --scope userCheck both boundaries
`omx doctor` catches missing OMX files, hooks and prerequisites; the exec smoke test catches auth, profile and provider/base-URL problems that only appear on a real request.
omx doctor
codex login status
omx exec --skip-git-repo-check -C . "Reply with exactly OMX-EXEC-OK"Launch the recommended way
Run from the git project you want Codex to edit, with a branch-like task name. This starts the leader in an OMX-managed detached tmux so the HUD and runtime panes can be created and recovered.
omx --worktree=feat/task --madmax --xhighWork inside Codex
Set a durable objective, clarify, approve the plan, then wrap it in durable goals.
/goal Create a safe authentication refactor plan, implement it, and verify login, logout, and refresh-token behavior.
$deep-interview "clarify the authentication change"
$ralplan "approve the auth plan and review tradeoffs"
$ultragoal "turn the approved plan into durable Codex goals"Run concurrent conversations
Each additional conversation needs its own explicit root — reusing one is a fatal owner conflict by design.
omx
OMX_ROOT="$HOME/.omx/instances/second-conversation" omxCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Run long refactors through a repeatable clarify → plan → execute → verify loop instead of ad-hoc Codex prompts
- Keep risky bypass-mode sessions off your main checkout by launching each one into its own git worktree
- Split one story across coordinated parallel agents when a single-owner loop is too slow
- Carry plans, logs and mode state in-repo under `.omx/` so a session can be picked up later or by someone else
How oh-my-codex (OMX) compares
oh-my-codex (OMX) alongside other open-source ai sdlc automation tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| Multica | ★ 50.4k | Self-hostable workspace where coding agents are assigned issues like teammates: 26 agent CLIs, runtimes you own, a replayable execution log per run, and review gates before anything ships. |
| GPT-Pilot | ★ 33.7k | Autonomous AI developer that breaks down an app description into tasks, writes and runs code incrementally, and asks clarifying questions to produce a working production application. |
| oh-my-codex (OMX) | ★ 33.2k | A workflow layer that makes OpenAI Codex CLI sessions start stronger |
| Vibe Kanban | ★ 28.1k | Vibe Kanban lets you plan tasks on a kanban board, run coding agents like Claude Code and Codex in isolated workspaces, then review their diffs and ship pull requests. |
| Beads | ★ 27.3k | A Dolt-backed dependency-graph issue tracker for coding agents: hash IDs avoid multi-agent collisions, `bd ready` surfaces unblocked work, and `bd remember` keeps durable project memory. |
| Archon | ★ 23.5k | A workflow engine for AI coding agents: describe plan, implement, validate, review and PR phases as YAML, and every run repeats them in its own git worktree. |
| Keploy | ★ 18.5k | An API testing tool that records real traffic with eBPF — including database and queue calls — and replays it as deterministic tests and data mocks, with no SDK to import and no code changes. |
| Superset | ★ 14.4k | An agentic IDE for macOS that runs 100+ CLI coding agents in parallel, each in its own git worktree, with a built-in terminal, diff viewer, in-app browser and completion notifications. |