| name | omx |
| description | Use when you need multi-agent orchestration for OpenAI Codex CLI, especially Codex equivalents of Claude `/team`, `/autopilot`, `/ultrawork`, and `/ultraqa`. Triggers on: omx, $plan, $ralph, $team, $ulw, $ultraqa, $autopilot, $deep-interview. v0.11.10 — 30+ agents, 35+ workflow skills, tmux team runtime, sparkshell, explore, ralplan. |
| allowed-tools | Read Write Bash Grep Glob Edit Agent |
| metadata | {"tags":"omx, oh-my-codex, codex, openai, multi-agent, orchestration, team, autopilot, ultrawork, ultraqa, ralph, plan, deep-interview, ralplan","platforms":"Codex, Claude, Gemini","keyword":"omx","version":"0.11.10","source":"Yeachan-Heo/oh-my-codex"} |
omx (oh-my-codex) — Multi-Agent Orchestration for Codex CLI
When to use this skill
- You want a stronger Codex CLI runtime with reusable role prompts and workflow skills
- You need coordinated parallel agents via tmux team workers
- You want Codex/OMX equivalents for Claude
/team, /autopilot, /ultrawork, or /ultraqa
- Complex tasks requiring persistent execution (
$ralph) or consensus planning ($ralplan)
- You need repository exploration (
omx explore) or bounded shell inspection (omx sparkshell)
- Cross-model consultation with
$ask-claude or $ask-gemini
1. Quick Start
npm install -g @openai/codex oh-my-codex
omx setup
omx --madmax --high
Inside Codex:
/prompts:architect "analyze the authentication flow"
$plan "map the safest implementation path"
Mental model: Codex does the agent work. OMX adds better prompts, reusable workflow skills, and a runtime layer around it. Start with omx --madmax --high and let the agent pull in $team, $ralph, or other workflows when the task needs them.
2. Agent Roles (30+)
| Tier | Agents |
|---|
| Core Development | architect · planner · executor · debugger · verifier · explore |
| Quality Assurance | style-reviewer · quality-reviewer · api-reviewer · security-reviewer · performance-reviewer |
| Domain Experts | dependency-expert · test-engineer · build-fixer · designer · writer · qa-tester |
| Product Strategy | product-manager · ux-researcher · product-analyst · information-architect |
Invoke via: /prompts:<agent-name> "task"
3. Workflow Skills (35+)
| Skill | Trigger | Description |
|---|
autopilot | $autopilot | Full autonomous pipeline — idea to working code |
ralph | $ralph | Persistent execution loop until verified complete |
ultrawork | $ulw, $ultrawork | Maximum parallelism across independent agents |
team | $team | N coordinated agents with tmux/worktree isolation |
plan | $plan | Strategic planning before implementation |
ralplan | $ralplan | Consensus-based iterative planning |
deep-interview | $deep-interview | Socratic clarification before planning or execution |
deepsearch | $deepsearch | Codebase-focused deep search |
analyze | $analyze | Evidence-driven causal analysis (trace methodology) |
tdd | $tdd | Test-first development (red-green-refactor) |
build-fix | $build-fix | Fix build errors, type errors, toolchain failures |
code-review | $code-review | Comprehensive multi-dimension code review |
security-review | $security-review | Security audit — vulnerabilities, trust boundaries |
ultraqa | $ultraqa | Maximum parallelism QA/review pass |
visual-verdict | $visual-verdict | UI/UX visual review with verdict |
4. Team Runtime
omx team 3:executor "fix all TypeScript errors"
omx team 2:security-reviewer,1:architect "audit auth module"
omx team status <team-name>
omx team resume <team-name>
omx team shutdown <team-name>
Inside Codex:
$team 3:executor "parallelize this refactor"
Requirements:
| Platform | Dependency |
|---|
| macOS | brew install tmux |
| Ubuntu/Debian | sudo apt install tmux |
| Fedora | sudo dnf install tmux |
| Windows (native) | winget install psmux |
| Windows (WSL2) | sudo apt install tmux |
5. Claude Workflow Parity
Read references/parallel-quality-workflows.md when the user asks to use Claude-style team, autopilot, ultrawork, or ultraqa from Codex.
| Claude / OMC intent | OMX command | Use when |
|---|
/team N:role "task" | $team N:role "task" or omx team N:role "task" | Coordinated work needs a shared task list and isolated workers |
/autopilot "task" or autopilot: task | $autopilot "task" | End-to-end idea → plan → implementation → verification should stay Codex-native |
/ultrawork "task" | $ulw "task" or $ultrawork "task" | Independent lanes can run with little synchronization |
/ultraqa "target" | $ultraqa "target" | QA/review should fan out across test, security, performance, or UI concerns |
Do not reuse .omc/state/ from Claude sessions. OMX state and scratch belong under .omx/.
For vague greenfield requests, run $deep-interview before $autopilot. For high-assurance completion loops, use $plan or $ralplan before $ralph.
6. CLI Commands
Core launch
omx
omx --madmax --high
omx --xhigh
Launch flags
| Flag | Description |
|---|
--yolo | Fast execution, minimal verification |
--high | High reasoning effort |
--xhigh | Extra-high reasoning effort |
--madmax | Bypass approvals/sandbox (--dangerously-bypass-approvals-and-sandbox) |
--dry-run | Preview without execution |
--verbose | Verbose output |
Operator commands
omx setup
omx doctor
omx doctor --team
omx status
omx cancel
omx hud --watch
omx reasoning <mode>
omx help
Explore & inspect
omx explore --prompt "find where team state is written"
omx sparkshell git status
omx sparkshell --tmux-pane %12 --tail-lines 400
Instructions
- Identify intent — match request to the best workflow skill (
$autopilot, $plan, $ralph, $team, $deep-interview)
- Start with the recommended launch —
omx --madmax --high for most sessions
- Use role prompts for analysis —
/prompts:architect "..." before implementation
- Escalate when needed — use
$autopilot for full autonomous build intent, $team for coordinated workers, $ulw for independent burst work, and $ultraqa for QA fan-out
- Use
$deep-interview for vague requests — when intent, scope, or boundaries are unclear
- Point to references — link to the relevant references/ file for deep-dive details
Examples
Example 1: Standard session
omx --madmax --high
/prompts:architect "analyze the current auth flow"
$plan "implement OAuth safely"
Example 2: Persistent execution
$ralph "fix all failing tests and make them pass"
Loop: execute → verify → fix, until all tests pass.
Example 3: Parallel team work
omx team 3:executor "refactor the API layer with proper tests"
Example 4: Claude ultrawork / ultraqa parity
$autopilot "build the task API end to end"
$ulw "parallelize the independent cleanup tasks"
$ultraqa "review the auth flow across tests, security, and performance"
Example 5: Clarify before coding
$deep-interview "I want to add a payment system"
One-question-at-a-time Socratic loop until intent is clear, then hands off to $plan or $ralph.
Best practices
- Codex-first — OMX is a workflow layer, not a replacement for Codex; start with
omx --madmax --high
- Use
$plan before $ralph — plan first when scope is unclear; ralph for guaranteed execution
$ralplan for consensus — use when multiple planning perspectives are needed before committing
$deep-interview gates execution — always use for vague greenfield or brownfield work
- Team mode is optional — use
$team for coordinated workers, $ulw for independent burst work, and $ultraqa for QA fan-out
- Cross-model consultation —
$ask-claude and $ask-gemini for second opinions on design or architecture
omx explore for safe lookups — read-only; won't modify anything
omx doctor when stuck — diagnose installation issues before manual debugging
Quick Reference
| Command/Skill | Action |
|---|
omx --madmax --high | Recommended launch |
/prompts:architect "..." | Analysis and design review |
$autopilot "..." | Full autonomous idea-to-working-code pipeline |
$plan "..." | Strategic planning |
$deep-interview "..." | Clarify vague requests |
$ralph "..." | Persistent loop until done |
$ralplan "..." | Consensus planning |
$team N:role "..." | Parallel coordinated agents |
$ulw "..." | Maximum parallelism |
$ultraqa "..." | Parallel QA/review burst |
omx explore --prompt "..." | Read-only repo exploration |
omx sparkshell <cmd> | Bounded shell inspection |
omx setup | Install/re-install OMX |
omx doctor | Diagnose issues |
omx cancel / $cancel | Cancel active modes |
→ Website · GitHub · npm
References