oh-my-claudecode

A Claude Code plugin for staged multi-agent orchestration, skills, native teams, and optionally launched CLI workers.

oh-my-claudecode (OMC) is an open-source Claude Code plugin by Yeachan Heo and contributors. Its official product and documentation entry point is [yeachan-heo.github.io/oh-my-claudecode-website](https://yeachan-heo.github.io/oh-my-claudecode-website); source, release history, manifests, and the security guide are in the [Yeachan-Heo/oh-my-claudecode repository](https://github.com/Yeachan-Heo/oh-my-claudecode). OMC does not add a separate hosted chat product to Claude Code. Instead, it packages local plugin components: skills, commands, agent roles, hooks, a local MCP bridge process, and an npm-based CLI runtime. Its aim is to structure planning, delegation, verification, and repair as repeatable workflows inside a Claude Code environment.

Verified as a Claude Code plugin

The primary source makes the classification explicit. .claude-plugin/marketplace.json declares the omc marketplace, names Yeachan Heo as owner, and lists oh-my-claudecode as a plugin sourced from ./ in the productivity category. The adjacent .claude-plugin/plugin.json calls it a “Multi-agent orchestration system for Claude Code,” points to the repository, and specifies the MIT license. The manifest supplies skills, commands, and .mcp.json; the latter starts the bundled bridge/mcp-server.cjs locally through Node. This is why the record belongs to the Plugin catalog type and the existing Coding category: its primary outcome is coordinated software work, not a separate general-purpose cloud service.

The catalog assigns OMC only the Claude Code compatibility. That is the documented plugin surface and Marketplace installation route. OMC can additionally call external, already installed and authenticated CLIs in terminal mode: its README names Claude, Codex, Gemini, Antigravity, Grok, and Cursor for omc ask and Team workers. That does not mean the Claude Code plugin is automatically installable in those products. The source specifically directs Codex users to the separate oh-my-codex project. Teams should preserve that distinction and separately approve, install, and authenticate every external tool.

Installation and practical workflows

The provider-recommended route consists of two Claude Code slash commands entered one at a time:

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

Then /omc-setup configures the plugin environment in an active Claude Code session. The repository also documents an alternative terminal runtime through npm i -g oh-my-claude-sisyphus@latest, followed by omc setup; the npm package name is intentionally different from the product and plugin name. The session surface includes /team, /autopilot, /execute, /ralph, /ralplan, /deep-interview, /verify, and /ask. Per the README, Team is the canonical staged orchestration workflow, while Autopilot, Execute, and Ralph target different levels of automation, persistence, and verification loops. These commands do not guarantee that any task is completed correctly: requirements, permissions, models, the working directory, and project quality still determine the outcome.

Data, network, and security boundaries

According to the README, OMC writes runtime state such as sessions, plans, logs, handoffs, research notes, and local artifacts under .omc/ by default. Those files can contain prompts, transcripts, or machine-specific paths and should not be committed or copied into tickets without review. In worktrees this state normally lives inside the individual worktree; OMC_STATE_DIR can change the location. The included MCP bridge starts locally, while the security guide identifies optional remote MCP services for Exa web search and Context7 documentation. Team and Ask workflows can also use external providers and their local CLIs. When those are enabled, task material, tool results, or prompts may reach the relevant CLI, API, or MCP path. Before use, review privacy, retention, access control, and permitted data types for the service actually selected.

The provider documents OMC_SECURITY=strict as a conservative mode: it restricts AST tools to the project root, enables a Python REPL blocklist safeguard, disables remote MCP, blocks external LLM workers, prevents silent auto-updates, and caps persistent loops. That is useful but not a complete sandbox. The security guide explicitly identifies the absence of OS-level process isolation and the absence of a security boundary between parallel agents. One especially important point is that external CLI workers may run unattended with broad auto-approval flags. Start with a non-sensitive repository, least privilege, a reviewed tool and network policy, and human review of every change. Credentials do not belong in prompts, .omc artifacts, or unprotected environment variables.

License, popularity, and a suitable rollout

The repository is MIT licensed. GitHub’s official page metadata showed exactly 39,033 stars on 2026-09-07; the catalog stores that repository-wide, point-in-time number, not a measurement of the plugin’s standalone adoption. Stars are not evidence of quality, privacy, or security. The source describes a freely available open-source distribution, but this catalog entry gives no concrete prices: Claude Code, selected models, API access, external CLIs, or infrastructure may create costs. Check current terms with the relevant providers.

OMC suits teams and individuals doing recurring development work who want planning, parallel specialist roles, and visible quality checks organized more consistently. It is less suitable for tightly isolated environments where an external CLI or an extra MCP process is unacceptable, or for tasks whose risk prohibits automated file-writing or execution access. A reliable first rollout installs the plugin in a test project, reviews configuration created by /omc-setup, runs a small task with explicit acceptance criteria, inspects .omc and the Git diff, and only then deliberately enables additional workers or network paths.

Installation instructions

oh-my-claudecode, or OMC for short, is an open-source plugin for Claude Code that, according to the provider, promises multi-agent orchestration with zero learning curve. The project is maintained by Yeachan Heo as creator and lead and is available in the Yeachan-Heo/oh-my-claudecode GitHub repository under the MIT license, complemented by a dedicated documentation website. The core idea is to extend Claude Code with specialized subagents, intelligent model routing, and reusable skills, so that, per the project description, users do not first have to learn how Claude Code works in detail but can simply use OMC.

Step 1: Installing via marketplace or npm

For most users, the provider recommends installing through the Claude Code plugin marketplace: first the marketplace is added using the repository path as a slash command inside Claude Code, then the actual plugin is installed. Both lines must be entered individually according to the instructions, since pasting both commands at once fails. Alternatively, an npm package called oh-my-claude-sisyphus is available for global installation, offering a standalone CLI runtime path. The provider notes a known, harmless npm warning about a deprecated dependency of a native SQLite extension that currently cannot be fixed without risking the installation itself.

Step 2: Setup and first steps

After installation, setup runs either as a slash command inside an active Claude Code session or as a standalone terminal command. If OMC is launched through a custom plugin path rather than the regular marketplace mechanism, a plugin-directory mode must additionally be enabled, or a corresponding environment variable set, so that setup does not accidentally duplicate skills and agents the plugin already provides at runtime. Initial tasks can then be kicked off through an autopilot command that accepts a natural-language task description and automatically breaks it into planning, execution, and quality-assurance phases, optionally using named workflow profiles stored in a configuration file.

Orchestration, agents, and model routing

According to the provider, OMC ships nineteen specialized agents with different tiers for architecture, research, design, testing, and data analysis, and automatically routes tasks to the appropriate underlying model, for example a smaller, faster model for simple tasks and a more capable model for complex reasoning. A HUD status line shows real-time metrics about the ongoing orchestration, and a skill-learning mechanism extracts reusable patterns from past sessions, which are then automatically surfaced in the future whenever they match the current context.

Project and user skills plus state management

Custom skills learned from sessions can be stored either at the project level in a hidden folder inside the repository or at the user level in the home directory, with project-level skills taking priority over user-level ones and being shareable with a team when committed to the repository. According to the documentation, OMC writes runtime state, session data, plans, logs, and research notes into its own hidden folder; this is excluded from version control by default, with the explicit exception of the skills subfolder, which is deliberately meant to remain shareable. For several independent repositories sharing a common parent working directory, a marker file can be used to set up a shared state root.

Practical value and limitations

The practical value is that complex, multi-step coding tasks no longer need to be manually broken down into individual prompts but are automatically coordinated through an orchestration layer with matching specialist agents and model selection, which can save time especially on larger refactors or feature builds. The limitation is that OMC is a third-party extension built on top of Claude Code and assumes its base behavior and API access; anyone who does not understand the orchestration logic risks automatically selected agents or models not fitting their actual task, which is why the compatibility matrix from the documentation is worth checking before production use.

Installation

/plugin install oh-my-claudecode
Free
Provider
Yeachan Heo
License
MIT
Last reviewed
07.09.2026
GitHub stars
39,038

Categories

Compatible with

Claude Code