oh-my-codex

Fri, 03 Apr 2026 16:06:43 +0800

Yeachan-Heo/oh-my-codex

oh-my-codex (OMX)

Multi-agent orchestration layer for OpenAI Codex CLI. Inspired by oh-my-claudecode.

What is this?

OMX brings 30 specialized agent roles, 39 workflow skills, mode lifecycle management, team orchestration, verification protocols, and notification systems to Codex CLI – without forking the upstream project.

Architecture

User -> Codex CLI -> AGENTS.md (orchestration brain)
                  -> ~/.codex/prompts/*.md (30 agent slash commands)
                  -> ~/.agents/skills/*/SKILL.md (39 skills)
                  -> config.toml (MCP servers, notify hook, features)
                  -> .omx/ (state, memory, notepad, plans)

Key design decision: OMX is an add-on, not a fork. It uses Codex CLI’s native extension points:

Extension Point	What OMX Uses It For
`AGENTS.md`	Orchestration brain (equivalent to CLAUDE.md)
`~/.codex/prompts/*.md`	30 agent definitions as slash commands
`~/.agents/skills/*/SKILL.md`	39 workflow skills
`config.toml` MCP servers	State management + project memory
`config.toml` notify	Post-turn logging and metrics
`config.toml` features	collab (sub-agents), child_agents_md

Quick Start

# Install
npm install -g oh-my-codex

# Setup (installs prompts, skills, configures Codex CLI)
omx setup

# Verify installation
omx doctor

# Start using in Codex CLI
codex
> /architect "analyze the auth module"
> /autopilot "build a REST API for user management"
> /team 3:executor "fix all TypeScript errors"

Setup Details

omx setup performs these steps:

Creates directories (~/.codex/prompts/, ~/.agents/skills/, .omx/state/)
Installs 30 agent prompt files to ~/.codex/prompts/
Installs 39 skill SKILL.md files to ~/.agents/skills/
Updates ~/.codex/config.toml with MCP servers and features
Generates AGENTS.md in the current project root
Configures the notify hook for post-turn logging

Agent Catalog (30 agents)

Build & Analysis

Agent	Model	Description
`/explore`	haiku	Codebase discovery, symbol/file mapping
`/analyst`	opus	Requirements clarity, acceptance criteria
`/planner`	opus	Task sequencing, execution plans
`/architect`	opus	System design, boundaries, interfaces
`/debugger`	sonnet	Root-cause analysis, failure diagnosis
`/executor`	sonnet	Code implementation, refactoring
`/deep-executor`	opus	Complex autonomous goal-oriented tasks
`/verifier`	sonnet	Completion evidence, claim validation

Review

Agent	Model	Description
`/style-reviewer`	haiku	Formatting, naming, lint conventions
`/quality-reviewer`	sonnet	Logic defects, anti-patterns
`/api-reviewer`	sonnet	API contracts, versioning
`/security-reviewer`	sonnet	Vulnerabilities, OWASP Top 10
`/performance-reviewer`	sonnet	Hotspots, complexity optimization
`/code-reviewer`	opus	Comprehensive review across concerns

Domain Specialists

Agent	Model	Description
`/dependency-expert`	sonnet	External SDK/API evaluation
`/test-engineer`	sonnet	Test strategy, coverage
`/quality-strategist`	sonnet	Release readiness, risk assessment
`/build-fixer`	sonnet	Build/toolchain failures
`/designer`	sonnet	UX/UI architecture
`/writer`	haiku	Docs, migration notes
`/qa-tester`	sonnet	Interactive CLI validation
`/scientist`	sonnet	Data/statistical analysis
`/git-master`	sonnet	Commit strategy, history hygiene

Product

Agent	Model	Description
`/product-manager`	sonnet	Problem framing, PRDs
`/ux-researcher`	sonnet	Heuristic audits, usability
`/information-architect`	sonnet	Taxonomy, navigation
`/product-analyst`	sonnet	Product metrics, experiments

Coordination

Agent	Model	Description
`/critic`	opus	Plan/design critical challenge
`/vision`	sonnet	Image/screenshot analysis

Skills (39 skills)

Execution Modes

autopilot - Full autonomous execution from idea to working code
ralph - Persistence loop with architect verification
ultrawork - Maximum parallelism with parallel agent orchestration
team - N coordinated agents on shared task list
pipeline - Sequential agent chaining with data passing
ecomode - Token-efficient execution using haiku and sonnet
ultrapilot - Parallel autopilot with file ownership partitioning
ultraqa - QA cycling: test, verify, fix, repeat
swarm - Compatibility facade over team

Planning

plan - Strategic planning with optional consensus/review modes
ralplan - Consensus planning (planner + architect + critic)

Agent Shortcuts

analyze - Deep analysis and investigation
deepsearch - Thorough codebase search
tdd - Test-driven development workflow
build-fix - Fix build and TypeScript errors
code-review - Comprehensive code review
security-review - Security vulnerability review
frontend-ui-ux - UI/UX design work
git-master - Git operations
review - Plan review/critique

Utilities

cancel - Cancel any active mode
doctor - Diagnose installation issues
help - Usage guide
note - Save notes to notepad
trace - Agent flow trace timeline
skill - Manage local skills
learner - Extract learned skills
research - Parallel research agents
deepinit - Deep codebase init with AGENTS.md
release - Automated release workflow
hud - Status display configuration
omx-setup - Setup and configure OMX
configure-telegram - Telegram notifications
configure-discord - Discord notifications
writer-memory - Writer memory system
project-session-manager / psm - Isolated dev environments
ralph-init - Initialize PRD for ralph-loop
learn-about-omx - OMX usage patterns

MCP Servers

OMX provides two MCP servers that Codex CLI connects to via config.toml:

State Server (`omx_state`)

Manages mode lifecycle state for all execution modes.

state_read / state_write / state_clear
state_list_active / state_get_status
Supports: autopilot, ralph, ultrawork, ecomode, ultrapilot, team, pipeline, ultraqa, ralplan

Memory Server (`omx_memory`)

Project memory and session notepad.

project_memory_read / project_memory_write
project_memory_add_note / project_memory_add_directive
notepad_read / notepad_write_priority / notepad_write_working / notepad_write_manual

Team Orchestration

The team skill provides a staged pipeline:

`1`	`team-plan -> team-prd -> team-exec -> team-verify -> team-fix (loop)`

Each stage uses specialized agents. The verify/fix loop is bounded by max attempts to prevent infinite loops. Terminal states: complete, failed, cancelled.

Notification System

Supports three channels (configured in .omx/notifications.json):

Desktop - Native notifications (macOS, Linux, Windows)
Discord - Webhook integration with embedded messages
Telegram - Bot API with Markdown formatting

Verification Protocol

Evidence-backed verification with task sizing:

Small (<5 files, <100 lines): Type check + related tests
Standard: Full type check + test suite + linter
Large (>20 files): Full check + security review + performance assessment + regression testing

Hook Emulation

Codex CLI has limited hook support (AfterAgent, AfterToolUse only). OMX emulates the full hook pipeline:

OMC Hook	OMX Equivalent	Capability
SessionStart	AGENTS.md native loading	Full
PreToolUse	AGENTS.md inline guidance	Partial
PostToolUse	notify config hook	Partial
UserPromptSubmit	AGENTS.md self-detection	Partial
SubagentStart/Stop	Codex CLI collab native	Full
Stop	notify config	Full

Coverage

~92% feature parity with oh-my-claudecode (excluding MCP tools).

See COVERAGE.md for the detailed feature matrix.

Project Structure

oh-my-codex/
  bin/omx.js              # CLI entry point
  src/
    cli/                   # CLI commands (setup, doctor, version, status, cancel, help)
    config/                # config.toml generator
    agents/                # Agent definitions registry
    mcp/                   # MCP servers (state, memory)
    hooks/                 # Hook emulation layer + keyword detector
    modes/                 # Mode lifecycle management
    team/                  # Team orchestration (staged pipeline)
    verification/          # Verification protocol
    notifications/         # Desktop/Discord/Telegram notifications
    utils/                 # Path resolution, package utilities
  prompts/                 # 30 agent prompt files (*.md)
  skills/                  # 39 skill directories (*/SKILL.md)
  templates/               # AGENTS.md template
  scripts/                 # notify-hook.js
  dist/                    # Compiled output

Development

# Install dependencies
npm install

# Build
npm run build

# Type check
npx tsc --noEmit

# Link for local development
npm link

License

MIT

Workflow Skills on Producthunt daily