oh-my-claudecode

Multi-agent orchestration for Claude Code. Run teams of AI agents in parallel — each with a role, shared context, and coordinated workflows. Zero learning curve.

GitHub: Yeachan-Heo/oh-my-claudecode

Overview

oh-my-claudecode (OMC) lets you run teams of Claude Code agents in parallel, each with assigned roles. It manages a staged pipeline (plan → PRD → execute → verify → fix) and supports mixing providers (Claude, Codex, Gemini). Agents share context through the filesystem and git branches, with automatic conflict resolution.

Instructions

Installation

Via Claude Code Plugin Marketplace (recommended):

/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
/plugin install oh-my-claudecode

Via npm:

npm i -g oh-my-claude-sisyphus@latest

Initial Setup

Run inside Claude Code:

/setup
/omc-setup

Enable experimental teams in ~/.claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Team Pipeline

Team mode runs a staged pipeline for every task:

team-plan → team-prd → team-exec → team-verify → team-fix (loop)

1. Plan — break down the task into sub-tasks and dependencies
2. PRD — generate a product requirements document
3. Execute — spawn N agents working in parallel
4. Verify — validate output against requirements
5. Fix — loop back to fix issues until tests pass

Agent Roles

Role	Focus	Example Tasks
executor	General coding	Feature implementation, refactoring
reviewer	Code review	PR reviews, architecture feedback
tester	Quality assurance	Test writing, coverage analysis
architect	System design	API design, database schema

Context Sharing

Agents in a team share context through:

• Shared filesystem — all agents see the same project files
• Team state file — .omc/team-state.json tracks progress
• Git branches — each agent works on a feature branch, merged at verify stage

Configuration

~/.omc/config.json:

{
  "defaultTeamSize": 3,
  "defaultRole": "executor",
  "providers": {
    "claude": { "enabled": true },
    "codex": { "enabled": true, "model": "codex-latest" },
    "gemini": { "enabled": true, "model": "gemini-2.5-pro" }
  },
  "pipeline": {
    "skipPRD": false,
    "autoFix": true,
    "maxFixLoops": 3
  }
}

Requires codex / gemini CLIs installed and an active tmux session for multi-provider workers (v4.4.0+).

Examples

Example 1: Team Execution with Mixed Roles

# Spawn 3 executor agents to fix all TypeScript errors
/team 3:executor "fix all TypeScript errors"

# Use Codex agents for code review
omc team 2:codex "review auth module for security issues"

# Gemini agents for UI work
omc team 2:gemini "redesign UI components for accessibility"

Example 2: Autopilot and Deep Interview

# Let OMC handle everything — from planning to execution
autopilot: build a REST API for managing tasks

# When requirements are vague, use Socratic questioning first
/deep-interview "I want to build a task management app"
# The interview clarifies requirements before any code is written

Example 3: Tri-Model Advisor

Route work to Codex + Gemini, then Claude synthesizes the results:

/ccg Review this PR — architecture (Codex) and UI components (Gemini)

Guidelines

• Start with autopilot: for well-defined tasks
• Use /deep-interview when requirements are fuzzy
• Mix providers: Codex for review, Gemini for UI, Claude for logic
• Keep team size at 5 or fewer for most tasks — diminishing returns beyond that
• Use omc team status <task-id> to monitor long-running teams
• The verify-fix loop catches most issues automatically
• When multiple agents modify the same file, OMC detects conflicts at merge time and spawns a resolver agent

Resources

• Documentation
• CLI Reference
• Workflows
• Discord
• Migration Guide