Turning Claude Code into an AI Command Center with 28 Agents and 8 Execution Modes

What is oh‑my‑claudecode (OMC)?

OMC is a multi‑agent orchestration plugin for Claude Code that transforms the tool from a single‑purpose assistant into a coordinated AI development team. It provides 28 specialized agents, 8 execution modes, 37 built‑in skills, and IDE‑level LSP/AST tooling.

Problems addressed

Efficiency – Native Claude Code processes tasks sequentially, causing long wait times for refactoring, testing, and documentation.

Cost – Claude API charges per token; without routing users must manually choose between cheap (Haiku) and expensive (Opus) models.

Quality – A single agent handling all tasks can miss issues; OMC adds dedicated reviewers, security checkers, and QA testers.

Installation

Prerequisites: Claude Code CLI installed and a Claude Max/Pro subscription or an Anthropic API key.

# Step 1: Add plugin marketplace
/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode

# Step 2: Install the plugin
/plugin install oh-my-claudecode

# Step 3: Run initialization
/omc-setup

# Verify installation
/omc-doctor   # expects "All checks passed"

Agent architecture

Agents are grouped into three tiers, each mapped to an Anthropic model:

LOW (Haiku) – quick lookups and simple tasks. Example agents: explore, executor‑low, designer‑low.

MEDIUM (Sonnet) – standard implementations and moderate reasoning. Example agents: executor, designer, qa‑tester.

HIGH (Opus) – complex reasoning and architecture design. Example agents: architect, planner, code‑reviewer.

Roles cover analysis, execution, design, documentation, testing, security review, and planning, forming a virtual development team.

Execution modes

Eight modes allow different trade‑offs between coordination and parallelism:

Team (recommended) – staged pipeline (plan → prd → exec → verify → fix) with coordinated agents.

Autopilot – fully autonomous single‑agent flow for end‑to‑end feature development.

Ultrawork – maximum parallelism for bulk fixes or large‑scale refactoring.

Ralph – persistent execution with automatic retry until completion.

Pipeline – strict sequential processing for multi‑step transformations.

omc‑teams – real CLI workers (Claude/Codex/Gemini) for Codex/Gemini tasks.

ccg – parallel Codex (analysis) and Gemini (design) for backend + UI work.

Swarm/Ultrapilot – compatibility layer routing to Team mode.

Smart routing

OMC evaluates task complexity and selects the optimal model, reducing token cost by 30‑50%.

# Routes to Haiku (simple)
"find all files that import react"

# Routes to Sonnet (standard)
"implement a login form with validation"

# Routes to Opus (complex)
"review the entire codebase for security issues"

Skill system

OMC extracts reusable patterns from interactions into built‑in skills such as git‑master, tdd, code‑review, pr‑generator, refactor. Invoke a skill by prefixing the command:

tdd: implement password validation

This runs the test‑engineer agent, which follows the RED‑GREEN‑REFACTOR TDD cycle.

LSP and AST tools

OMC adds IDE‑level intelligence with 12 LSP commands and 2 AST commands. Supported languages include TypeScript, Python, Rust, Go, C/C++, Java, JavaScript, Ruby, Kotlin, and Swift.

# LSP commands
lsp_hover
lsp_goto_definition
lsp_find_references
lsp_document_symbols
lsp_workspace_symbols
lsp_diagnostics
lsp_diagnostics_directory

# AST commands
ast_grep_search
ast_grep_replace

Practical scenarios

Rapid prototyping – autopilot: build a todo app with React splits the task into design, front‑end, back‑end, and testing agents.

Large‑scale refactor – ralph ulw: refactor the entire API layer runs multiple executors in parallel, verifies each change, and retries failures automatically.

Code review – /oh‑my‑claudecode:code‑review activates the Opus‑based code‑reviewer, which produces a structured audit report.

Multi‑AI collaboration –

/ccg Review this PR — architecture (Codex) and UI components (Gemini)

runs Codex for architecture analysis and Gemini for UI review, with Claude aggregating the results.

Comparison with native Claude Code

Agent count: basic vs 28 specialized agents.

Execution modes: single vs 8 modes.

Model routing: manual vs smart automatic.

Parallel execution: limited vs full support.

LSP tools: none vs 12 tools.

AST tools: none vs 2 tools.

Persistent execution: none vs Ralph mode.

Learning system: none vs skill extraction.

Cost optimization: none vs 30‑50% token savings.

Multi‑AI orchestration: none vs Claude + Codex + Gemini.

Best practices

Enable Agent Teams by setting "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS":"1" in ~/.claude/settings.json.

Use magic keywords ( ralph, ulw, plan, ralplan) to control behavior precisely.

Configure HUD ( /hud setup) for real‑time visibility of agent activity.

Prefer project‑level .claude/CLAUDE.md over global settings.

Update regularly with /plugin marketplace update omc then /omc-setup.

Pitfalls

Package name on npm is oh‑my‑claude‑sisyphus while the brand is oh‑my‑claudecode.

Version 4.4.0 removed Codex/Gemini MCP servers; use /omc‑teams instead.

Old -low/-medium/-high suffixes are deprecated; specify the model via the model parameter.

On Windows use Node.js hooks (.mjs) instead of .sh scripts.

API rate limits can be mitigated with omc wait, which backs off automatically.

Use /omc‑doctor for diagnosing installation issues.

Pros and cons

Pros

Zero learning curve – natural‑language interaction.

Powerful orchestration: 28 agents, 8 modes, 37 skills.

Smart cost optimization saves 30‑50% tokens.

IDE‑level intelligence via LSP and AST tools.

Persistent execution ensures task completion.

Supports multiple AI models (Claude, Codex, Gemini).

Real‑time HUD provides visibility.

Skill learning extracts reusable patterns.

Active maintenance (over 170 releases).

Cons

Depends on Claude Code CLI.

Requires Claude Max/Pro subscription or API key.

Feature richness leads to a steep learning curve for newcomers.

Package name mismatch can cause confusion.

Frequent version changes require migration attention.

Additional subscriptions needed for Codex/Gemini.

Resources

GitHub repository: https://github.com/Yeachan-Heo/oh-my-claudecode

Architecture documentation: https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/docs/ARCHITECTURE.md

Reference manual: https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/docs/REFERENCE.md

Migration guide: https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/docs/MIGRATION.md