oh-my-codex 0.1.3

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

Oh My Codex

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

Installation • Quick Start • Modes • Agents • Tools • 한국어

---

Why Oh My Codex?

Codex CLI is powerful alone. Oh My Codex makes it a team.

Codex CLI	Oh My Codex
Single agent	32 specialized agents
Manual model selection	Auto model routing
Sequential execution	Parallel execution
No session memory	Session persistence

Based on Vercel's research: AGENTS.md achieves 100% pass rate vs 79% for skills-only.

Installation

Option 1: PyPI (Recommended)

pip install oh-my-codex
omx-setup  # Interactive setup wizard

Option 2: From Source

git clone https://github.com/junghwaYang/oh-my-codex.git
cd oh-my-codex
./install.sh

Requirements

• Python 3.10+
• Codex CLI installed
• OpenAI API key or Codex Pro subscription

Quick Start

# Basic usage (direct Codex)
omx "fix the bug in auth.py"

# Multi-agent autonomous execution
omx "autopilot: build a REST API with auth"

# Parallel execution
omx "ulw: refactor all components to TypeScript"

# Never give up mode
omx "ralph: fix all failing tests"

Execution Modes

Keyword	Mode	Description	Agents
autopilot:	Autopilot	Full autonomous execution	PM → Executor → Tester → Reviewer
ulw:	Ultrawork	Parallel multi-agent	PM + Frontend + Backend + Tester
ultrapilot:	Ultrapilot	Maximum parallelism	All specialists
team:	Team	Pipeline orchestration	plan → exec → verify → fix loop
ralph:	Ralph	Persistent (never gives up)	PM → Executor → Debugger
plan:	Plan	Planning only, no execution	Planner + Architect
eco:	Eco	Token-efficient, fast	Single Executor
tdd:	TDD	Test-driven development	Tester → Executor
review:	Review	Code review	Reviewer + Security
debug:	Debug	Systematic debugging	Debugger + Analyst

Examples

# Complex feature development
omx "autopilot: implement OAuth2 with Google and GitHub"

# Parallel refactoring
omx "ulw: convert all class components to hooks"

# Persistent bug fixing
omx "ralph: the login redirect isn't working, fix it"

# Architecture planning
omx "plan: design a microservices architecture"

# Quick fix (token-efficient)
omx "eco: add error handling to api calls"

# Test-driven development
omx "tdd: implement password validation"

# Security review
omx "review: audit the authentication module"

Models & Reasoning

Auto Model Selection

Task Complexity	Model
Trivial	gpt-5-nano
Simple	gpt-5-mini
Standard	gpt-5.1-codex
Complex	gpt-5.2-codex
Long-running	gpt-5.1-codex-max

Reasoning Effort

Level	Usage	Auto-mapped Modes
none	Fast responses	eco
low	Light tasks	tdd, pipeline
medium	Balanced	plan, ultrawork
high	Deep thinking	autopilot, review
xhigh	Maximum (5.2-codex)	ralph, ultrapilot, debug

# Manual override
omx --reasoning xhigh "complex architecture decision"
omx --model gpt-5.2-codex "critical task"

Agents (32)

Orchestration

Agent	Model	Role
PM	gpt-5.2-codex	Master orchestrator, delegates tasks
Coordinator	gpt-5.2-codex	Manages parallel execution
Executor	gpt-5.1-codex	Gets things done

Development

Agent	Expertise
Frontend	React, Vue, TypeScript, CSS
Backend	Node.js, Python, APIs, DBs
Fullstack	End-to-end development
Mobile	React Native, Flutter
DevOps	CI/CD, Docker, K8s

Quality

Agent	Focus
Tester	Unit, integration, E2E tests
Reviewer	Code review, best practices
Security	Vulnerabilities, OWASP
Debugger	Systematic bug hunting

Specialized

Agent	Domain
Architect	System design, patterns
Researcher	Information gathering
Data/ML	Data engineering, ML
Writer	Documentation

Tools (9)

Agents have access to these tools:

Tool	Description
run_shell	Execute terminal commands
read_file	Read file contents
write_file	Create/overwrite files
edit_file	Precise text replacement
list_directory	Browse directories
search_files	Find files by name/content
git_status	Check git state
git_diff	View changes
run_tests	Auto-detect & run tests

Skills (31)

Installed to ~/.codex/skills/:

Category	Skills
Orchestration	team, autopilot, ultrawork, ultrapilot, ralph, pipeline, swarm
Planning	planner, ralplan, analyze, research, deepsearch
Development	eco, tdd, build-fix, deepinit, release
Quality	reviewer, code-review, security-review, ultraqa
Utilities	git-master, playwright, debug, mcp-setup, doctor, hud, trace

CLI Reference

# Basic
omx "task"                      # Auto-detect mode
omx "autopilot: task"           # Explicit mode

# Options
omx --model gpt-5.2-codex "task"  # Model override
omx --reasoning high "task"       # Reasoning level
omx --provider openai "task"      # Use API billing
omx -v "task"                     # Verbose output

# Sessions
omx --list                      # List all sessions
omx --resume <session_id>       # Resume session
omx --status                    # Show current config

# Setup
omx-setup                       # Run setup wizard
omx --set-provider codex        # Change billing

Configuration

~/.codex/config.toml

[model]
default = "gpt-5.1-codex"

[model.routing]
nano = "gpt-5-nano"
mini = "gpt-5-mini"
standard = "gpt-5.1-codex"
powerful = "gpt-5.2-codex"
max = "gpt-5.1-codex-max"

[model.reasoning]
default = "none"
autopilot = "high"
ralph = "xhigh"
eco = "none"

[billing]
provider = "codex"  # or "openai"

[skills]
auto_load = true

Architecture

User: omx "autopilot: build API"
         │
         ▼
    ┌─────────────┐
    │ Mode Detect │ → autopilot
    │ Model Route │ → gpt-5.2-codex
    │ Reasoning   │ → high
    └──────┬──────┘
           ▼
    ┌─────────────┐
    │ PM Agent    │ ← tools: shell, files, git
    │ + Handoffs  │ → [Executor, Tester, Reviewer]
    └──────┬──────┘
           ▼
    ┌─────────────┐
    │ Runner.run  │ ← OpenAI Agents SDK
    │ Autonomous  │
    └──────┬──────┘
           ▼
       Results

Comparison

Feature	Codex CLI	omx
Single agent	✅	✅
Multi-agent	❌	✅ 32 agents
Parallel execution	❌	✅ ultrawork
Auto model routing	❌	✅
Reasoning control	Manual	✅ Auto-mapped
Session persistence	❌	✅
Skills	✅	✅ 31 included

Credits

• oh-my-claudecode — Original inspiration
• OpenAI Agents SDK — Multi-agent framework
• Vercel Research — AGENTS.md philosophy

License

MIT © junghwaYang