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
• 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

/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+)

architect

planner

executor

debugger

verifier

explore

style-reviewer

quality-reviewer

api-reviewer

security-reviewer

performance-reviewer

dependency-expert

test-engineer

build-fixer

designer

writer

qa-tester

product-manager

ux-researcher

product-analyst

information-architect

/prompts:<agent-name> "task"

3. Workflow Skills (35+)

autopilot

$autopilot

ralph

$ralph

ultrawork

$ulw

team

$team

plan

$plan

ralplan

$ralplan

deep-interview

$deep-interview

deepsearch

$deepsearch

analyze

$analyze

tdd

$tdd

build-fix

$build-fix

code-review

$code-review

security-review

$security-review

ultraqa

$ultraqa

visual-verdict

$visual-verdict

frontend-ui-ux

$frontend-ui-ux

ai-slop-cleaner

$ai-slop-cleaner

ask-claude

$ask-claude

ask-gemini

$ask-gemini

git-master

$git-master

ecomode

$ecomode

web-clone

$web-clone

cancel

$cancel

4. Team Runtime

# Start parallel team workers
omx team 3:executor "fix all TypeScript errors"
omx team 2:security-reviewer,1:architect "audit auth module"

# Team management
omx team status <team-name>
omx team resume <team-name>
omx team shutdown <team-name>

$team 3:executor "parallelize this refactor"

brew install tmux

sudo apt install tmux

sudo dnf install tmux

winget install psmux

sudo apt install tmux

5. CLI Commands

Core launch

omx                   # Launch Codex with OMX runtime
omx --madmax --high   # Recommended default (bypass + high reasoning)
omx --xhigh           # Extra-high reasoning effort

Launch flags

--yolo

--high

--xhigh

--madmax

--dangerously-bypass-approvals-and-sandbox

--dry-run

--verbose

Operator commands

omx setup             # Install prompts, skills, config, AGENTS scaffolding
omx doctor            # Verify installation
omx doctor --team     # Team/swarm diagnostics
omx status            # Show active modes
omx cancel            # Cancel execution modes
omx hud --watch       # Live HUD status monitor
omx reasoning <mode>  # Set: low/medium/high/xhigh
omx help              # Show help

Explore & inspect

omx explore --prompt "find where team state is written"   # Read-only repo lookup
omx sparkshell git status                                  # Bounded shell inspection
omx sparkshell --tmux-pane %12 --tail-lines 400           # Tail tmux pane output

Instructions

1. Identify intent — match request to the best workflow skill (

   $plan

   ,

   $ralph

   ,

   $team

   ,

   $deep-interview

   )
2. Start with the recommended launch —

   omx --madmax --high

   for most sessions
3. Use role prompts for analysis —

   /prompts:architect "..."

   before implementation
4. Escalate when needed — let the agent pull in heavier workflows; don't force

   $team

   for simple tasks
5. Use

   $deep-interview

   for vague requests — when intent, scope, or boundaries are unclear
6. 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"

Example 3: Parallel team work

omx team 3:executor "refactor the API layer with proper tests"

Example 4: Clarify before coding

$deep-interview "I want to add a payment system"

$plan

$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 — only escalate to

  $team

  when the task genuinely needs parallel workers
• 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

omx --madmax --high

/prompts:architect "..."

$plan "..."

$deep-interview "..."

$ralph "..."

$ralplan "..."

$team N:role "..."

$ulw "..."

omx explore --prompt "..."

omx sparkshell <cmd>

omx setup

omx doctor

omx cancel

$cancel

References