If you’ve been using Claude Code for a while, you’ve probably already noticed the pattern: it works great on isolated tasks, but coordinating complex, multi-step work still requires a lot of manual steering. You write a prompt, review the output, write another prompt, redirect, repeat.
oh-my-claudecode (OMC) is a plugin that tries to eliminate that friction. It wraps Claude Code with a multi-agent orchestration layer — 32 specialized agents, 40+ skills, and five distinct execution modes — all accessible in natural language. No new syntax to memorize, no configuration required.
15.4K stars on GitHub, 27K downloads per month, MIT license. Worth understanding how it works.
What exactly is OMC?
The name is a nod to oh-my-zsh: a framework that installs on top of an existing tool and makes it significantly more capable without replacing it. OMC installs as a Claude Code plugin and turns Claude Code into what the project calls a “conductor” — an orchestrator that delegates work to specialized agents instead of doing everything itself.
The core idea: instead of a single Claude instance handling everything sequentially, OMC divides work into roles (architect, executor, qa-tester, analyst, designer, and 27 more) and routes each task to the right agent at the appropriate model tier — Haiku for simple tasks, Sonnet for intermediate work, Opus for complex reasoning. The result is faster execution and lower token cost.
Installation
# From inside Claude Code:
/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
/plugin install oh-my-claudecode
# Then run setup:
/oh-my-claudecode:omc-setup
Done. OMC reads your project, auto-configures itself, and injects the agents into your Claude Code session. No configuration files to manually edit.
If you prefer npm directly:
npm install -g oh-my-claude-sisyphus
Note: the npm package is called oh-my-claude-sisyphus — a reference to the Greek myth, which makes sense when you see how the persistence modes work.
The 5 execution modes
This is where OMC sets itself apart from simpler plugins. You don’t pick a mode from a menu — you describe what you want in natural language and OMC activates the appropriate mode. But understanding what each mode does helps you frame your requests better.
1. Autopilot — fully autonomous execution
autopilot: build a REST API for managing tasks
Autopilot takes a high-level description and orchestrates the complete cycle — exploration, planning, implementation, testing — without asking you for intermediate decisions. It delegates to specialized agents, routes to the correct model tier, and verifies with the Architect agent before finishing.
Best for: features from scratch, new modules, tasks where you want to describe the outcome and review at the end.
2. Ultrapilot — maximum parallelism
ultrawork implement user authentication with OAuth
Ultrapilot runs up to 5 concurrent workers simultaneously. Independent tasks that would normally run sequentially are parallelized. The declared speedup is 3-5x compared to sequential execution.
Best for: large refactors, test suite generation, migrations — work that can be decomposed into parallel streams.
3. Swarm — coordinated team over a shared task list
team 3:executor "build a fullstack todo app"
Swarm spins up N coordinated agents working from a shared task list, using Claude Code’s native agent teams feature. Agents take tasks, complete them, and the list advances. Real-time messaging and integrated task coordination.
Best for: projects that need real multi-agent coordination with visibility into who’s doing what.
4. Pipeline — sequential execution by stages
The pipeline mode runs work through structured stages: plan → PRD → implementation → verification → correction loop. Each stage produces a handoff artifact that the next stage consumes. If verification fails, it automatically loops back.
Best for: features that benefit from explicit planning before implementation — when you’ve already been burned by Claude diving into code before understanding requirements well.
5. Ecomode — token-efficient execution
eco: refactor the authentication module
Ecomode activates token-saving strategies — agents load on-demand, simple tasks route to Haiku, context is managed more aggressively. The project declares a 30-50% token savings compared to standard execution.
Best for: routine work where cost matters, or when you’re working in long sessions and want to stretch your usage.
Skills: portable debugging knowledge
Beyond execution modes, OMC includes a skills system. Skills are structured knowledge files that automatically inject when relevant:
# .omc/skills/fix-proxy-crash.md
---
name: Fix Proxy Crash
description: aiohttp proxy crashes on ClientDisconnectedError
triggers: ["proxy", "aiohttp", "disconnected"]
---
When OMC detects a problem related to proxies, it loads this skill automatically and the agent has the context it needs without you explaining it again. You build your own library of hard-won debugging knowledge that persists across sessions.
Some things worth knowing
Ralph mode is integrated. If you used Ralph Loop separately, OMC includes its own persistence mode (prefix ralph:) that keeps running until a task completes — the same Sisyphus concept, built in directly.
Cross-validation with Codex and Gemini is optional. OMC can route verification steps to external models. This requires separate CLI access to those tools, but is available if you want multi-model review.
The HUD gives you real-time observability. A status line shows active agents, token usage, and session state — useful for long autonomous runs where you want visibility without interrupting execution.
Windows support is experimental. OMC requires tmux. WSL2 works; native Windows has limited support.
When to use OMC vs Claude Code directly
OMC adds value when your task is complex enough to benefit from agent delegation. For quick questions, simple edits, or short implementations, standard Claude Code is faster to get started.
A rough heuristic:
| Scenario | Recommended mode |
|---|---|
| Single file change or quick question | Claude Code directly |
| Multi-file feature with clear requirements | Autopilot |
| Large parallel work load | Ultrapilot |
| Real coordination between multiple agents | Swarm |
| Need to plan before coding | Pipeline |
| Routine work, cost-sensitive | Ecomode |
Conclusion
oh-my-claudecode is one of the most mature plugins in the Claude Code ecosystem. The five execution modes cover distinct real-world scenarios, the skills system solves a real pain point (context disappearing between sessions), and zero-config setup means you can evaluate it in five minutes.
15K+ stars and 27K monthly downloads suggest the community has already done that evaluation. Worth adding to your Claude Code setup and running Autopilot on your next feature to see how far it goes on its own.
Have you tried OMC, or are you using a different orchestration approach with Claude Code? Share your setup in the comments — especially if you’ve compared it with Superpowers or Ralph Loop 
Links: