oh‑my‑claudecode

What it is

• A multi‑agent orchestration tool that sits on top of Claude Code (Anthropic’s code‑focused LLM assistant) — it lets you run tasks like building a REST‑API, reviewing code or refactoring with natural‑language commands.
• It provides two surfaces:
  1. In‑session slash‑commands (/omc‑setup … / /autopilot … / /team … …) that you type inside a Claude Code chat‑window. 2 Terminal‑based CLI (omc …) that talks to the same orchestration engine and can be used in scripts or CI pipelines.
• The project is published on npm as oh‑my‑claude‑sisyphus and also as a Claude Code marketplace plugin.

---

Quick start (the 3‑step flow the README shows)

1. Install
   • Preferred for Claude Code users: add the plugin from the marketplace and run /plugin install oh‑my‑claudecode.
   • Or install the CLI globally: npm i -g oh‑my‑claude‑sisyphus@latest.
2. Set up

   /omc-setup          # inside a Claude Code chat
   # or from the shell
   omc setup
   

   This creates the configuration files (.claude/omc.jsonc or ~/.config/claude‑omc/config.jsonc) and registers the built‑in skills.
3. Run a task

   /autopilot "build a REST API for managing tasks"
   

   The command is parsed, a team of Claude agents is spawned, they plan, implement, test and verify the feature, and you see a live HUD‑like status line.

---

Core concepts & terminology

Term	Meaning
Autopilot	The default autonomous workflow that runs a single lead‑agent through the stages plan → execution → verify (optionally with a QA step). It is the simplest way to ask Claude to “just do it”.
Team	A staged pipeline (team‑plan → team‑prd → team‑exec → team‑verify → team‑fix) that can run multiple agents in parallel using tmux panes. Each pane runs a real CLI version of a provider (Claude, Codex, Gemini, Antigravity, Grok, Cursor).
/ask	A shortcut to query a specific provider (e.g. codex or antigravity) and let Claude synth‑​​ize the answers.
Workflow / profile	A JSON‑C configuration (.claude/omc.jsonc) that defines which stages to run (ralplan, execution, qa, …) and optional model routing.
HUD	A small status line shown in the Claude Code UI that reports the current stage, active agents, token usage and any errors.
Plugin directory flags	Options (--plugin-dir, --plugin-dir-mode) that tell the installer where the plugin lives so it does not duplicate skills when the runtime already provides them.

---

Main features

• Zero‑configuration start – install and run /omc‑setup; defaults work for most projects.
• Team‑first orchestration – the recommended mode; parallelises work across several LLM‑agents.
• Natural‑language interface – you only need to describe the goal; no memorised CLI syntax.
• Cost‑aware routing – the engine can pick cheaper models for certain stages, saving 30‑50 % of token spend.\n- Persistent execution – the system keeps trying until the verification checks (tests, lint, type‑check) pass.
• Pattern reuse – completed tasks are stored as “artifacts” that can be re‑used in future runs.
• Real‑time HUD – shows progress, token usage, and any failures while the agents work.

---

Who is it for?

• Developers who want to automate repetitive coding tasks (scaffolding, refactoring, code‑review) without writing custom scripts.
• Teams that already use Claude Code and would like a higher‑level orchestration layer to run multi‑agent pipelines.
• CI/CD pipelines that need a head‑less, deterministic CLI (omc …) to invoke Claude‑driven actions.

---

Limitations & gotchas

• The advanced Team mode requires Linux (the flock utility) and a running tmux session.
• Some commands exist only inside a Claude Code chat (/autopilot, /team) and are not available as CLI sub‑commands.
• The npm install pulls in a native better‑sqlite3 dependency that prints a deprecation warning (prebuild‑install@7.1.3); it does not affect functionality.
• To use the CLI workers you must have the provider‑specific CLIs installed and authenticated (e.g., codex, gemini, cursor‑agent).

---

How to keep it up‑to‑date

• npm install: npm i -g oh‑my‑claude‑suffix… (the README calls it sisyphus).
• Marketplace: /plugin marketplace update omc then /omc-setup.
• Run /omc‑doctor if the plugin cache becomes stale.

---

Bottom line

oh‑my‑claudecode is a genuine, actively‑maintained open‑source project that adds a multi‑agent orchestration layer on top of Anthropic’s Claude Code. It is useful for developers who want to issue high‑level natural‑language commands and let a coordinated set of LLM agents plan, write, test and verify code, either from within Claude Code or from a terminal/CI environment.