oh-my-codex 0.1.0

Multi-agent orchestration for OpenAI Codex CLI - Turn Codex into a team of AI agents

Oh My Codex

Multi-agent orchestration system for OpenAI Codex CLI

Features • Quick Start • Modes • Skills • Multi-Agent • 한국어

---

Inspired by oh-my-claudecode, adapted for OpenAI Codex CLI.

Why Oh My Codex?

OpenAI Codex CLI is powerful, but lacks the multi-agent orchestration that makes Claude Code + oh-my-claudecode so effective. This project bridges that gap.

Based on Vercel's research showing AGENTS.md outperforms skills (100% vs 79%), we use an AGENTS.md-first design.

Features

Feature	Description
🧠 AGENTS.md-First	Core orchestration always in context
🚀 8 Execution Modes	team, autopilot, ultrawork, ralph, pipeline, eco, plan, ultrapilot
🔧 31 Native Skills	Full toolkit matching oh-my-claudecode
🤖 32 Specialized Agents	From PM to Data Scientist
📊 Smart Routing	Automatic model selection
💾 Session Management	Pause, resume, track
📡 HUD & Tracing	Real-time metrics and debugging

Quick Start

# Clone
git clone https://github.com/junghwaYang/oh-my-codex.git
cd oh-my-codex

# Install
./install.sh

# Use
omx "autopilot: build a REST API for tasks"

Execution Modes

Mode	Keyword	Description
Team	team:	Canonical multi-agent pipeline (plan→exec→verify→fix)
Autopilot	autopilot:	Full autonomous execution
Ultrawork	ulw:	Parallel multi-file operations
Ralph	ralph:	Persistent mode (never gives up)
Ultrapilot	ultrapilot:	Maximum parallelism
Pipeline	pipeline:	Sequential staged processing
Eco	eco:	Token-efficient execution
Plan	plan:	Interview-driven planning

Examples

# Team orchestration (recommended for complex tasks)
omx "team: build a fullstack app with auth"

# Autopilot for feature development
omx "autopilot: implement user dashboard"

# Parallel refactoring
omx "ulw: rename userId to user_id everywhere"

# Persistent debugging
omx "ralph: fix all TypeScript errors"

# Token-efficient quick fix
omx "eco: add .env to gitignore"

# Planning without execution
omx "plan: design the payment system"

Skills (31)

Orchestration

Skill	Description
team	Multi-agent staged pipeline
autopilot	Autonomous execution
ultrawork	Parallel execution
ultrapilot	Maximum parallelism
ralph	Persistent mode
pipeline	Sequential processing
swarm	Legacy multi-agent (→ team)

Planning & Analysis

Skill	Description
planner	Interview-driven planning
ralplan	Iterative planning consensus
analyze	Code quality analysis
research	Deep research
deepsearch	Codebase exploration

Development

Skill	Description
eco	Token-efficient mode
tdd	Test-driven development
build-fix	Fix build errors
deepinit	Project initialization
release	Version & changelog

Quality & Review

Skill	Description
reviewer	Code review
code-review	Comprehensive review
security-review	Security audit
ultraqa	Parallel testing

Tools & Utilities

Skill	Description
git-master	Git workflows
playwright	E2E testing
debug	Systematic debugging
mcp-setup	MCP configuration
configure-notifications	Alerts setup

System

Skill	Description
doctor	Installation diagnostics
hud	Real-time metrics
trace	Execution tracing
learner	Pattern extraction
note	Session notes

Agents (32)

Primary Orchestration

• PM — Master orchestrator
• Coordinator — Parallel execution management
• Executor — Task execution
• Deep Executor — Complex implementations

Planning & Analysis

• Planner — Creates actionable plans
• Analyst — System analysis
• Researcher — Information gathering
• Explorer — Codebase navigation

Architecture & Design

• Architect — System design
• Designer — UI/UX design
• System Designer — Distributed systems

Development

• Frontend — React, Vue, TypeScript
• Backend — APIs, databases
• Fullstack — End-to-end development
• Mobile — React Native, Flutter
• DevOps — CI/CD, infrastructure

Quality & Testing

• Tester — Unit/integration tests
• QA — Quality assurance
• Security — Security engineering
• Performance — Performance optimization

Review & Critique

• Reviewer — Code review
• Critic — Challenge assumptions

Specialized

• Scientist — Data science
• Data — Data engineering
• ML — Machine learning
• Writer — Documentation
• Docs — API documentation
• Vision — Visual analysis

Support

• Debugger — Bug finding
• Refactorer — Code improvement
• Migrator — Upgrades & migrations

Architecture

oh-my-codex/
├── AGENTS.md                 # Core orchestration brain
├── .codex/skills/            # 31 native skills
│   ├── team/
│   ├── autopilot/
│   ├── ultrawork/
│   └── ... (28 more)
├── orchestrator/             # Python multi-agent
│   ├── agents/               # 32 agent definitions
│   ├── session.py            # Session management
│   ├── mcp.py                # MCP servers
│   ├── router.py             # Model routing
│   └── cli.py                # CLI entry
├── bin/omx                   # CLI wrapper
├── config.toml               # Configuration
└── install.sh                # Installer

CLI Reference

omx "task description"          # Auto-detect mode
omx "autopilot: task"           # Explicit mode
omx -m ultrawork "task"         # Force mode
omx --model gpt-4.1 "task"      # Model override
omx --list                      # List sessions
omx --resume <id>               # Resume session
omx --status                    # Check status
omx -v "task"                   # Verbose

Configuration

~/.codex/config.toml

[model]
default = "o3"

[model.routing]
simple = "gpt-4.1-mini"
standard = "gpt-4.1"
complex = "o3"

[hud]
enabled = true
style = "standard"

[trace]
enabled = false
level = "standard"

Credits

• oh-my-claudecode — Original inspiration
• oh-my-opencode — Architecture reference
• Supabase Agent Skills — Skill structure
• Vercel Research — Design philosophy

License

MIT