Master Oh My Claude Code: Complete Guide to Multi‑Agent AI Coding with Claude

What Is Oh My Claude Code

Oh My Claude Code (OMC) is a collection of plugins and agents built on Claude Code that expands the single‑point interaction model into a multi‑agent orchestration system. It aggregates ideas from oh‑my‑opencode , claude‑hud , Superpowers and everything‑claude‑code to provide three core capabilities: multiple execution modes, a suite of 32 specialized agents covering common development roles, and minimal configuration to start immediately.

Core Capabilities

5 execution modes : Autopilot, Ultrapilot (3–5× parallel), Swarm, Pipeline, Ecomode.

32 professional agents ranging from architecture analysis to UI design.

Automatic model routing : simple tasks use Haiku, routine work uses Sonnet, complex reasoning falls back to Opus.

Natural‑language control : describe the goal and OMC selects the appropriate mode and agents.

Installation & Configuration

All steps are performed inside Claude Code:

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

/plugin install oh-my-claudecode

Run the setup wizard to generate a .claude/CLAUDE.md configuration file:

/oh-my-claudecode:omc-setup

Project‑Level vs Global Configuration

Project‑level (recommended) creates .claude/CLAUDE.md in the current repository, affecting only that project: /oh-my-claudecode:omc-setup --local Global configuration writes to ~/.claude/CLAUDE.md and applies to every Claude Code session. When both exist, the project‑level file takes precedence.

Verification

After setup, confirm OMC is active with a simple command: autopilot: create a simple hello world function If OMC is correctly installed, Claude replies that it has taken over the execution flow.

Five Execution Modes in Detail

1. Autopilot – Fully autonomous mode

Autopilot is the flagship mode. The user states the desired outcome and OMC handles planning, implementation, testing, and self‑correction until the task is complete.

Automatic requirement gathering and information collection.

Parallel execution by specialized agents.

Continuous verification and testing.

Self‑correction loops.

Typical use cases: building a complete feature module or a small application without manual intervention.

2. Ultrapilot – Parallel‑accelerated mode (3–5× speedup)

Ultrapilot extends Autopilot by spawning up to five parallel workers to split a task into concurrent subtasks.

/oh-my-claudecode:ultrapilot "build a fullstack todo app"

Automatic task decomposition for parallel execution.

File‑ownership coordination to avoid conflicts.

Overall speed increase of roughly 3–5× for multi‑component projects.

Best for large‑scale refactoring, multi‑module systems, or any project where development time must be minimized.

3. Swarm – Collaborative team mode

Swarm creates a pool of agents that each claim the smallest granularity of work, complete it, and then pick the next task.

/oh-my-claudecode:swarm 5:executor "fix all TypeScript errors"

Atomic task splitting to avoid duplicate work.

5‑minute timeout per task with automatic release.

Supports 2–10 parallel workers.

Process ends naturally when the task pool is empty.

Ideal for bulk independent issues (e.g., fixing many TypeScript errors) or batch code modifications.

4. Pipeline – Sequential pipeline mode

Pipeline strings together multiple agents in a fixed order, passing each stage’s output directly to the next stage.

/oh-my-claudecode:pipeline explore:haiku -> architect:opus -> executor:sonnet

Built‑in pipelines include review (explore → architect → critic → executor), implement (planner → executor → tdd‑guide), debug (explore → architect → build‑fixer) and refactor (explore → architect‑medium → executor‑high → qa‑tester). These cover common development workflows that require strict step ordering.

5. Ecomode – Economic mode

Ecomode reduces token consumption without noticeably sacrificing performance. It still supports parallel execution but automatically selects cheaper models for simple subtasks.

/oh-my-claudecode:ecomode "refactor the authentication system"

Simple, mechanical tasks → Haiku.

General development work → Sonnet.

Complex reasoning → Opus.

This mode is especially friendly to Pro users who need to balance speed and token budget.

Magic Keywords & Natural‑Language Delegation

OMC infers the appropriate mode and agents from natural‑language descriptions. For precise control, magic keywords can explicitly specify modes and behaviors. Example combining keywords: ralph ulw: migrate the entire database ralph – enables persistence so the task won’t stop midway.

ulw – activates Ultrapilot parallelism for faster execution.

Intelligent Model Routing

OMC automatically routes tasks to the most suitable model based on complexity, saving tokens while maintaining quality. Simple tasks avoid the expensive Opus model, whereas complex tasks receive the full reasoning power of Opus.

32 Professional Agent Ecosystem

Agents are grouped by role. Each agent is bound to a specific model tier (Opus, Sonnet, Haiku) that matches its expected workload.

Analysis & Architecture

architect (Opus) – complex debugging and system design.

architect‑medium (Sonnet) – regular analysis.

architect‑low (Haiku) – quick problem locating.

Execution

executor‑high (Opus) – complex refactoring.

executor (Sonnet) – feature implementation.

executor‑low (Haiku) – simple code tweaks.

Search & Exploration

explore‑high (Opus) – architecture‑level search.

explore‑medium (Sonnet) – deep code lookup.

explore (Haiku) – fast pinpointing.

Frontend & Design

designer‑high (Opus) – complex UI systems.

designer (Sonnet) – component‑level UI.

designer‑low (Haiku) – simple style adjustments.

Other Roles

planner (Opus) – task and solution planning.

critic (Opus) – solution review.

researcher (Sonnet) – documentation and data lookup.

writer (Haiku) – technical writing.

qa‑tester (Sonnet) – CLI testing.

scientist (Sonnet) – data analysis.

vision (Sonnet) – image‑related tasks.

When a user says “fix UI colors”, the task is routed to design‑related agents; when asking “why does this crash”, it goes to architecture and debugging agents. All routing happens automatically without manual specification.

References

[1] oh‑my‑opencode: https://github.com/code-yeongyu/oh-my-opencode

[2] claude‑hud: https://github.com/jarrodwatts/claude-hud

[3] Superpowers: https://github.com/NexTechFusion/Superpowers

[4] everything‑claude‑code: https://github.com/affaan-m/everything-claude-code

GitHub repository:

https://github.com/Yeachan-Heo/oh-my-claudecode