oh-my-claudecode: Turn Claude Code Into a Full 32-Agent Development Team

Claude Code has transformed how developers write code. With an estimated 4% of GitHub commits generated by Claude Code and a significant contribution to Anthropic's revenue, the tool has established itself as the most widely used development agent on the market. But it has a fundamental limitation: it is a single agent that works alone, sequentially.

oh-my-claudecode (OMC) is an open-source plugin that transforms Claude Code into a full team of 32 specialized agents. Architect, executor, designer, QA tester, security auditor, data scientist: each agent has a precise role and a model tier matched to the complexity of its task. The project, created by Yeachan Heo, has accumulated 13,400 GitHub stars, 877 forks, and 203 releases under the MIT license.

The idea is simple: why use a single generalist agent when you can orchestrate a team of specialists working in parallel?

To understand OMC's appeal, you first need to appreciate the scale of Claude Code's explosion in the developer ecosystem. Anthropic, which raised $30 billion in February 2026 at a $380 billion valuation, views Claude Code as a major strategic product. The tool is used by tens of thousands of developers daily, and some analyses estimate it generates up to 4% of all commits on GitHub.

But Claude Code, in its native version, remains fundamentally a single agent. You give it an instruction, and it executes sequentially. For a bug fix or a small feature, this is fine. For massive refactoring, architecture migration, or developing a complete feature with tests and documentation, sequential execution becomes a bottleneck.

Anthropic is working on native Agent Teams features (still experimental), but the community did not wait. OMC is the most comprehensive open-source answer to this need, and its 13,400 stars are proof of demand.

OMC organizes its 32 agents into three model tiers based on task complexity.

High-tier agents (Opus) handle tasks requiring deep reasoning: architect-high for system architecture design, scientist-high for complex data analysis, and critical code reviews.

Medium-tier agents (Sonnet) manage routine execution tasks: executor for code implementation, designer for interfaces, qa-tester for testing, security-reviewer for security audits.

Low-tier agents (Haiku) process simple and repetitive tasks: minor fixes, formatting, and maintenance operations that do not require complex reasoning.

This stratification is not just organizational. It has a direct impact on costs. By routing simple tasks to Haiku (Anthropic's cheapest model) and reserving Opus (the most powerful and expensive) for architectural decisions, OMC claims token savings of 30 to 50% compared to systematically using a single model.

The complete list includes agents for virtually every aspect of software development: documentation, refactoring, database migration, performance optimization, dependency management, and even API design.

OMC does more than provide agents. It offers multiple execution modes that determine how these agents collaborate.

Team mode is the most structured. It follows a canonical five-stage pipeline: planning, PRD (Product Requirements Document) writing, execution, verification, and correction. Each stage is assigned to the appropriate agent, and each stage's output feeds the next. This is the ideal approach for complex features that require upfront thinking.

The omc team mode uses tmux to launch parallel CLI workers. This approach allows not only parallelizing tasks across multiple Claude agents but also mixing agents from different providers: Claude, Codex, and Gemini can work simultaneously. Up to 5 concurrent workers are supported.

The ccg mode (tri-model advice) consults three different models on the same question and synthesizes their responses. It is an error-reduction mechanism through consensus, useful for critical technical decisions where a single viewpoint may be biased.

Autopilot mode lets the agent work autonomously on a task until completion, while Ralph mode maintains a persistent loop for long-running tasks.

Pipeline mode allows defining custom processing chains where one agent's output feeds the next agent's input according to a configurable graph.

Under the hood, OMC works through 31 lifecycle hooks that intercept different phases of Claude Code execution. These hooks allow injecting behavior before, during, and after command execution.

Among the most interesting hooks: autopilot manages autonomous execution, team-pipeline orchestrates teams, rules-injector injects contextual rules, and learner captures recurring patterns to transform them into reusable skills.

The skills system is particularly notable. OMC maintains a .omc/skills/ directory where learned patterns are stored and automatically injected into future contexts. If the architect identifies a recurring architecture pattern in your codebase, it is captured as a skill and automatically reused in future projects. The project includes 28+ built-in skills.

Notifications are supported via Telegram and Discord, and a monitoring HUD allows tracking agent activity in real time. For teams that use OMC intensively, these visibility features are essential for maintaining control over what the agents are doing.

Installation is done through the Claude Code plugin marketplace (/plugin) or via npm: npm i -g oh-my-claude-sisyphus (the project's historical npm name). The main prerequisite is tmux for multi-worker mode.

OMC advances significant performance figures, though it should be noted that no independent benchmarks have been published to validate them.

Parallel execution via Ultrapilot mode (an Autopilot variant) claims 3 to 5x acceleration over standard sequential execution. For massive refactoring or code review tasks across large codebases, this gain is plausible since multiple files can be processed simultaneously.

Token savings of 30 to 50% come from intelligent routing between model tiers. A simple calculation illustrates the impact: if 60% of your tasks are routed to Haiku instead of Sonnet, and Haiku costs roughly 10x less per token, the aggregate savings are substantial.

In terms of subscription costs, OMC works with existing Claude Code plans: Pro at $20/month, Max at $100-200/month. The plugin itself is free and open source (MIT). The real cost depends on usage intensity and the proportion of tasks routed to Opus versus Haiku.

The ecosystem is active: 2,000+ commits, 203 releases (the latest v4.9.1 dated March 24, 2026), and a growing contributor community.

The Claude Code plugin landscape is evolving rapidly, and OMC is not the only one targeting multi-agent orchestration.

TechDufus/oh-my-claude offers "ultrawork" prompts that optimize Claude Code performance without adding multiple agents. It is a lighter approach suited to users who want to stay close to the native Claude Code experience.

wshobson/agents offers more minimalist team orchestration, while huangdijia/oh-my-claude-code-plugins adds agents and MCP (Model Context Protocol) integrations.

Anthropic is also developing native Teams features in Claude Code (experimental at this stage), plus 28 official plugins including LSPs and GitHub integration. As these official features mature, the question will be whether third-party plugins like OMC remain necessary or get absorbed into the native product.

Outside the Claude ecosystem, alternatives like OpenCode, KiloCode, and RooCode offer their own multi-model approaches to AI-assisted development.

OMC's differentiation rests on three pillars: the number of specialized agents (32), tmux workers for real parallelism, and the skill learning system. It is the most comprehensive solution in the Claude Code ecosystem to date.

OMC targets a broad spectrum of developers but with different benefit levels depending on profile.

Solo developers who use Claude Code intensively (Max subscription) are the primary beneficiaries. Parallelism allows them to handle tasks that would take hours sequentially, and model-tier routing optimizes the cost of their subscription. For a freelance developer billing by the hour, a 3 to 5x gain on complex tasks translates directly into productivity and revenue.

Small development teams benefit from shared skills and hooks. When one developer captures an architectural pattern, it can be shared with the team through the .omc/skills/ directory, reducing onboarding time and standardizing practices.

Enterprises using Claude Code at scale can leverage multi-model mode (Claude, Codex, Gemini) to diversify their intelligence sources and reduce dependence on a single provider.

Limitations to keep in mind: OMC depends on your Claude subscription, the tmux configuration may require adjustment depending on your environment, and the learning curve for mastering the different execution modes is non-trivial. Starting with Team mode on a pilot project before moving to more advanced modes is the recommended approach.

The explosion of AI-assisted coding is just beginning. With Anthropic having raised $30 billion in February 2026 and Claude Code representing a growing share of its business, the plugin ecosystem will continue to expand. OMC, with its 13,400 stars and sustained development pace, is well positioned to remain a reference in this ecosystem.