Yeachan-Heo/oh-my-claudecode

A team-first multi-agent orchestration layer for Claude Code that adds autonomous execution modes, tmux-based CLI workers, and cross-provider advisors on top of the base tool.

What it is

oh-my-claudecode (OMC) is a multi-agent orchestration system for Claude Code that aims to let users describe what they want in natural language instead of learning Claude Code's mechanics. It exposes two surfaces: in-session slash commands and prompt triggers, and a terminal CLI (omc) that can spawn real worker processes in tmux panes. Its canonical orchestration mode is Team, a staged pipeline that plans, writes a PRD, executes, verifies, and loops on fixes. It targets developers who want coordinated, persistent, cost-aware agent workflows and optional cross-checking from other model CLIs.

Key features

Team mode as the canonical surface, running a staged pipeline (team-plan → team-prd → team-exec → team-verify → team-fix) with an in-session native variant and a tmux-CLI variant.
tmux CLI workers that launch real claude, codex, gemini, antigravity, grok, or cursor-agent processes on demand and shut them down when their task completes.
Multiple execution modes including Autopilot (autonomous single lead), Ultrawork (maximum parallelism), Ralph (persistent verify/fix loops), UltraQA (quality-gate cycling), and Pipeline.
19 specialized agents with tier variants and smart model routing between Haiku and Opus for cost optimization.
A provider advisor (omc ask / /ask) that runs local provider CLIs and saves markdown artifacts, plus a /ccg tri-model synthesis flow.
Custom, auto-injecting skills extracted from sessions, a HUD statusline, session/replay artifacts, and configurable notification callbacks for Telegram, Discord, and Slack.

Tech stack

TypeScript, targeting Node.js >=20.0.0, built with tsc and esbuild.
Depends on @anthropic-ai/claude-agent-sdk ^0.1.0 and @modelcontextprotocol/sdk ^1.26.0.
Uses better-sqlite3 for local state, commander for the CLI, zod/ajv for validation, @ast-grep/napi for structural code operations, and vscode-languageserver-protocol.
Tested with Vitest; distributed on npm as oh-my-claude-sisyphus (this excerpt is version 4.15.4), exposing both oh-my-claudecode and omc binaries.
Also exports TypeScript library helpers such as createOmcSession() for local Node.js programs. MIT licensed.

When to reach for it

You want Claude Code to coordinate multiple specialized agents through a staged, verified pipeline rather than issuing one-shot prompts.
You want autonomous or persistent execution that keeps working until a task is verified complete.
You want to cross-check work with Codex, Gemini, Grok, Antigravity, or Cursor CLIs from within one workflow.
You want built-in observability (HUD, session replays) and multi-repo workspace state management.

When not to reach for it

You do not use Claude Code; OMC is an orchestration layer for it, not a standalone agent.
You want interactive slash-command workflows in CI; the README advises against relying on /autopilot, /ralph, or /team in headless runs and to use deterministic omc commands instead.
You need the tmux CLI workers but cannot install or authenticate the underlying CLIs or run an active tmux session.
You prefer a minimal, low-churn setup; OMC is a large, fast-moving system with frequent releases and deprecations.

Maturity signal

Despite being created only in January 2026, the repository has nearly 38,000 stars and a most-recent push in July 2026 with just a single open issue, pointing to intense early adoption and very active, closely managed maintenance. The rapid version churn (this excerpt is v4.15.4) and repeated breaking changes documented in the README — removal of the swarm keyword in favor of Team, removal of the Codex/Gemini MCP servers in v4.4.0 — indicate a project still reshaping its core surfaces, so expect ongoing migration work. It is MIT licensed.

Alternatives

wshobson/agents — use it instead when you want a catalog of installable agents, skills, and commands rather than a runtime orchestration engine.
oh-my-codex — from the same author, use it instead when your primary harness is OpenAI Codex CLI rather than Claude Code.
Claude Code's native agent teams — use them instead when you want first-party team execution without an added orchestration layer.

Notes

The project's branding and the published npm package name diverge: the repo, plugin, and commands are branded oh-my-claudecode, but the npm package is oh-my-claude-sisyphus, which installs both the oh-my-claudecode and omc commands. The README also documents a deliberate architectural shift away from MCP servers toward CLI-first tmux workers, and it explicitly cautions against claiming that Claude Code's /goal evaluator independently runs commands or reads files.

Wiki: Yeachan-Heo/oh-my-claudecode