oh-my-claudecode Deployment & Tips: Double Your Efficiency with 32 Specialized Agents

Difficulty: Intermediate | Duration: 30 Minutes | Takeaways: Master multi-agent orchestration, zero-config deployment, and team pipeline setup.

Target Audience: Developers already using or planning to use Claude Code, and teams looking to further enhance Agent collaboration efficiency.

Core Dependencies: Node.js ≥18, tmux (cross-platform optional), Claude Code CLI

---

1. Project Introduction

Claude Code is powerful on its own, but a single Agent often struggles with complex multi-module projects—forcing you to repeatedly switch between conversation windows or manually coordinate code reviews, architectural design, and test validation.

oh-my-claudecode (hereafter referred to as OMC) is designed to solve this. It serves as a multi-agent orchestration layer for Claude Code, with the core philosophy:

Don't learn Claude Code. Just use OMC.

In short, you only need to describe what you want in natural language. OMC automatically schedules 32 specialized Agents, selects the appropriate models (Haiku / Sonnet / Opus), and processes tasks in parallel until verification passes. The entire process is zero-config and requires no memorization of commands.

Project URL: https://github.com/Yeachan-Heo/oh-my-claudecode

---

2. Core Features at a Glance

2.1 Zero-Config Out of the Box

No need to write YAML or JSON configurations. Once installed, just talk to it. Smart defaults cover 90% of scenarios.

2.2 32 Specialized Agents

Each Agent has a defined scope of responsibility. OMC automatically routes tasks to the most suitable Agent based on the task type:

The full list is defined in AGENTS.md.

2.3 Intelligent Model Routing

OMC automatically selects models based on task complexity:

• Haiku: Simple lookups, formatting, minor copy edits.
• Sonnet: Standard implementation, debugging, testing (daily workhorse).
• Opus: Architectural design, security reviews, complex refactoring.

This optimizes costs, with community feedback reporting an average of 30-50% savings in token consumption.

2.4 Automatic Parallelization

Independent tasks are automatically assigned to different Agents for parallel execution, without requiring manual decomposition.

2.5 Persistent Execution (verify/fix loop)

In Ralph mode, Agents automatically enter a verification loop after completing a task—if verification fails, they will self-fix and rerun until it truly passes.

2.6 Skill Self-Learning (Learner System)

Debugging experiences encountered during runtime are automatically extracted into reusable Skill files stored under .omc/skills/. Context is automatically injected when similar problems arise—no manual recall needed.

2.7 OpenClaw Integration

Supports forwarding Claude Code session events to the OpenClaw Gateway for automated notifications and remote triggers via channels like Telegram / Discord.

---

3. Full Project Structure

oh-my-claudecode/
├── agents/              # 32 specialized Agent definition files
├── bridge/              # OpenClaw / Third-party bridging
├── docs/                # Architecture docs, migration guides, performance monitoring
├── examples/            # Example workflows
├── hooks/               # Claude Code hook injection
├── missions/            # Mission definitions
├── research/            # Research modules
├── scripts/             # Utility scripts (OpenClaw Gateway Demo, etc.)
├── skills/              # Built-in OMC Skill files
├── src/                 # Core source code
├── tests/               # Test suites
├── CLAUDE.md            # Main Agent runtime config (Injected into Claude Code by OMC)
├── AGENTS.md            # Agent directory and descriptions
└── package.json

For most users, daily interactions involve:

• skills/ — Custom Skill files
• .omc/ — Runtime-generated session, state, and plan files

---

4. Installation & Configuration

4.1 Prerequisites

Required:

• Claude Code CLI — Requires Max/Pro subscription or Anthropic API Key
• Node.js ≥18 — OMC itself is a TypeScript/npm project

Optional but Highly Recommended:

• tmux — omc team and rate-limit auto-recovery depend on tmux

Install tmux for your platform:

# macOS
brew install tmux

# Ubuntu / Debian
sudo apt install tmux

# Arch
sudo pacman -S tmux

# Windows Native (Recommended)
winget install psmux

# Windows WSL2
sudo apt install tmux

4.2 Installing OMC (Two Methods)

Method 1: Marketplace Installation (Recommended)

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

# Install plugin
/plugin install oh-my-claudecode

Method 2: Global npm Installation

npm i -g oh-my-claude-sisyphus@latest

Once installed, OMC activates automatically; all / commands and magic keywords become available.

4.3 Initial Setup

# Run in any directory
/setup
/omc-setup

The first run will guide you through:

• Claude Code config directory confirmation
• OpenClaw Gateway (Optional)
• tmux session detection

If you encounter issues, run the diagnostic tool:

/omc-doctor

4.4 Enabling Team Mode (Recommended)

Team Mode is the default orchestration method since OMC v4.1.7. You need to enable experimental features in your Claude Code config:

# Edit ~/.claude/settings.json

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

---

5. Execution Modes Deep Dive

OMC provides various execution modes tailored for different scenarios. Understanding their differences is key to efficiency.

5.1 Team Mode (Recommended) ⭐

Best for: Complex multi-module tasks requiring architecture, coding, review, and testing collaboration.

Team Mode runs on a strict pipeline:

team-plan → team-prd → team-exec → team-verify → team-fix (loop)

The task starts with planning, moves through product design, code execution, verification, and returns to the execution phase if issues are found during the fix stage. OMC coordinates the entire process without manual intervention.

Basic Usage:

/team 3:executor "Fix all TypeScript compilation errors"

The number 3 indicates using 3 executor Agents for parallel processing.

CLI tmux Workers (v4.4.0+):

omc team 2:codex "Review auth module security"
omc team 2:gemini "Redesign UI component accessibility"
omc team 1:claude "Implement payment flow"
omc team status auth-review
omc team shutdown auth-review

These are true independent tmux processes that exit automatically upon completion, consuming no extra resources.

5.2 Autopilot

Best for: Clear requirements where a feature can be completed end-to-end without much human intervention.

autopilot: build a REST API for managing tasks

In Autopilot mode, a single Lead Agent handles everything from design to implementation without relying on a pipeline.

5.3 Ralph (Persistence Mode)

Best for: Tasks that must be completed fully, where "partial completion" is not acceptable.

Ralph mode includes the parallel power of Ultrawork plus the verify/fix loop:

executor → verify → fix → verify (until passed)

ralph: Refactor auth module, ensure all tests pass

If you find Claude Code often exiting when things "look finished" but aren't actually verified, Ralph is the solution.

5.4 Ultrawork (Maximum Parallelism)

Best for: Large-scale fixes or refactors requiring maximum speed.

ulw fix all ESLint errors
ulw rename all components to PascalCase

All fixes run in parallel without waiting for the previous one to finish—perfect for processing a "list of known issues."

5.5 /ccg (Triple-Model Advisor)

Best for: When you need perspectives from Codex and Gemini simultaneously, synthesized by Claude.

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

It calls /ask codex + /ask gemini, with Claude providing the final synthesized output. Ideal for full-stack reviews or cross-model verification.

5.6 Mode Comparison Table

---

6. Real-World Scenarios

6.1 Scenario 1: Building a REST API (Team Mode)

Suppose we want to build a Task Management REST API using Team Mode:

/team 2:executor "Implement a Task Management REST API with CRUD endpoints using Express + TypeScript"

OMC will automatically decompose the task:

1. architect designs API structure and data models.
2. executor implements endpoints in parallel.
3. test-engineer writes integration tests.
4. code-reviewer reviews the code.
5. verifier runs tests to validate.

You just wait for the final result.

6.2 Scenario 2: Auto-fixing All Errors (Ultrawork)

ulw fix all TypeScript type errors and ESLint warnings

Ideal for CI/CD pipelines or quickly cleaning up legacy projects.

6.3 Scenario 3: Unclear Requirements (Deep Interview)

If you aren't sure where to start:

/deep-interview "I want to build a personal blog system"

Deep Interview uses Socratic questioning to clarify needs: target audience, tech stack preferences, CMS requirements, SEO needs... clarify everything before starting to avoid rework.

6.4 Scenario 4: Cross-Model Review (/ccg)

/ccg Review the backend architecture with Codex and UI components with Gemini

6.5 Scenario 5: Rate Limit Auto-Recovery

When the Claude API triggers a rate limit:

omc wait          # View current status and recovery guidance
omc wait --start  # Start the auto-recovery daemon
omc wait --stop   # Stop the daemon

---

7. Troubleshooting

Q1: /team command is invalid, error "Team Mode not enabled"

Cause: CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS environment variable is not set.

Solution:

# Edit ~/.claude/settings.json and add:
{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Restart the Claude Code session for it to take effect.

---

Q2: omc team reports tmux-related errors

Cause: Windows user hasn't installed tmux, or tmux is not in the PATH.

Solution:

# Windows Native
winget install psmux

# macOS
brew install tmux

# Verify installation
tmux -V

If installed but still erroring, check if the psmux bin directory is in your system PATH.

---

Q3: OMC commands don't work after Marketplace installation

Cause: Plugin cache hasn't refreshed or the installation path is incorrect.

Solution:

# Update marketplace clone
/plugin marketplace update omc

# Re-initialize
/setup

# Run diagnostics
/omc-doctor

---

Q4: Claude Code shows rate limit but doesn't auto-recover

Cause: omc wait --start was not executed, or it's not running inside a tmux session.

Solution:

# Ensure execution within a tmux session
tmux new -s omc-watch
omc wait --start

The daemon runs in the tmux background and will automatically resume the Claude Code session once the limit resets.

---

Q5: Abnormal behavior after updating OMC

Cause: Old plugin cache is incompatible with the new code version.

Solution:

# For Marketplace installation:
/plugin marketplace update omc
/setup

# For npm installation:
npm i -g oh-my-claude-sisyphus@latest
/omc-doctor

---

Q6: Skill files are not being automatically injected

Cause: Skill trigger words don't match or the file path is wrong.

Solution:

Skill files are stored in two locations with different priorities:

# Project level (High priority): .omc/skills/
# User level (Low priority): ~/.omc/skills/

Ensure the triggers array in the Skill file frontmatter contains keywords from your current conversation:

---
name: Fix Proxy Crash
description: aiohttp proxy crashes on ClientDisconnectedError
triggers: ["proxy", "aiohttp", "disconnected"]  # Keyword matching
source: extracted
---

---

8. Further Reading / Advanced Directions

8.1 Writing Custom Skills

OMC supports extracting reusable Skills from actual debugging experiences:

/learner

This generates a YAML file under .omc/skills/. Context is automatically injected next time a similar issue occurs, eliminating manual recall.

8.2 OpenClaw Integration

OMC can forward Claude Code session events to the OpenClaw Gateway for remote triggers and notifications:

# Guided configuration
/oh-my-claudecode:configure-notifications
# → Select "OpenClaw Gateway"

Manually configure ~/.claude/omc_config.openclaw.json:

{
  "enabled": true,
  "gateways": {
    "my-gateway": {
      "url": "https://your-gateway.example.com/wake",
      "headers": { "Authorization": "Bearer YOUR_TOKEN" },
      "method": "POST",
      "timeout": 10000
    }
  },
  "hooks": {
    "session-start": { "gateway": "my-gateway", "instruction": "Session started for {{projectName}}", "enabled": true },
    "stop":          { "gateway": "my-gateway", "instruction": "Session stopping for {{projectName}}", "enabled": true }
  }
}

Supported hook events: session-start, stop, keyword-detector, ask-user-question, pre-tool-use, post-tool-use.

8.3 Notification System (Telegram / Discord / Slack)

omc config-stop-callback telegram --enable --token <bot_token> --chat <chat_id> --tag-list "@alice,bob"
omc config-stop-callback discord --enable --webhook <url> --tag-list "@here,123456789012345678"

Automatically push summaries to groups when a session ends, supporting incremental tag updates.

8.4 OpenClaw + OMC Collaborative Workflow

Combining OpenClaw’s automation with OMC’s multi-agent orchestration allows for:

• OpenClaw listening to Discord/Telegram messages to trigger a Claude Code session.
• OMC orchestrating 32 Agents within Claude Code to complete complex tasks.
• OpenClaw Gateway pushing the results back to the chat channel.

This gives you a 24/7 online multi-agent assistant that you can command through your daily chat tools.

See scripts/openclaw-gateway-demo.mjs to learn how to build this collaborative architecture.

8.5 Performance Monitoring & Debugging

OMC generates various files under .omc/ during runtime:

.omc/sessions/*.json        # Session summaries
.omc/state/agent-replay-*.jsonl  # Execution replay logs
.omc/plans/                 # Task plans
.omc/research/              # Research outputs

Run the HUD to view real-time orchestration status:

/oh-my-claudecode:hud setup
omc hud

---

Reference Links

• Project Homepage: https://github.com/Yeachan-Heo/oh-my-claudecode
• npm Package: https://www.npmjs.com/package/oh-my-claude-sisyphus
• Full Documentation: https://yeachan-heo.github.io/oh-my-claudecode-website
• CLI Reference: https://yeachan-heo.github.io/oh-my-claudecode-website/docs.html#cli-reference
• Discord Community: https://discord.gg/PUwSMR9XNk