oh-my-claudecode

Multi-agent orchestration for Claude Code. Zero learning curve.

Some advanced users customize zsh for years — most of us just use oh-my-zsh. Don't learn Claude Code. Just use OMC.

Get Started • Documentation • Migration Guide

---

Get Started (30 seconds)

Step 1: Install the plugin

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

Step 2: Run setup

/oh-my-claudecode:omc-setup

That's it. Everything else is automatic.

---

What Happens Now

When You...	I Automatically...
Give me a complex task	Parallelize with specialist agents
Say "plan this"	Start a planning interview
Say "don't stop until done"	Persist until verified complete
Work on UI/frontend	Activate design sensibility
Need research or exploration	Delegate to specialized agents
Say "build me..." or use autopilot	Execute full autonomous workflow

You don't need to memorize commands. I detect intent from natural language and activate the right behaviors automatically.

---

Magic Keywords (Optional Shortcuts)

These are optional shortcuts for power users who want explicit control. Natural language works just fine - these keywords simply provide precision when you want it.

Include these words anywhere in your message:

Keyword	Effect
ralph	Persistence mode - won't stop until done
ralplan	Iterative planning with consensus
ulw	Maximum parallel execution
plan	Start a planning interview
autopilot / ap	Full autonomous execution

Combine them: ralph ulw: migrate the database

---

Data Analysis & Research (v3.3.8)

Scientist Agent Tiers

Three tiers of scientist agents for quantitative analysis and data science:

Agent	Model	Use For
scientist-low	Haiku	Quick data inspection, simple statistics, file enumeration
scientist	Sonnet	Standard analysis, pattern detection, visualization
scientist-high	Opus	Complex reasoning, hypothesis validation, ML workflows

Features:

• Persistent Python REPL - Variables persist across calls (no pickle/reload overhead)
• Structured markers - [FINDING], [STAT:*], [DATA], [LIMITATION] for parsed output
• Quality gates - Every finding requires statistical evidence (CI, effect size, p-value)
• Auto-visualization - Charts saved to .omc/scientist/figures/
• Report generation - Markdown reports with embedded figures

# Variables persist across calls!
python_repl(action="execute", researchSessionID="analysis",
            code="import pandas as pd; df = pd.read_csv('data.csv')")

# df still exists - no need to reload
python_repl(action="execute", researchSessionID="analysis",
            code="print(df.describe())")

/oh-my-claudecode:research Command (NEW)

Orchestrate parallel scientist agents for comprehensive research workflows:

/oh-my-claudecode:research <goal>                    # Standard research with checkpoints
/oh-my-claudecode:research AUTO: <goal>              # Fully autonomous until complete
/oh-my-claudecode:research status                    # Check current session
/oh-my-claudecode:research resume                    # Resume interrupted session
/oh-my-claudecode:research list                      # List all sessions
/oh-my-claudecode:research report <session-id>       # Generate report for session

Research Protocol:

1. Decomposition - Breaks goal into 3-7 independent stages
2. Parallel Execution - Fires scientist agents concurrently (max 5)
3. Cross-Validation - Verifies consistency across findings
4. Synthesis - Generates comprehensive markdown report

Smart Model Routing:

• Data gathering tasks → scientist-low (Haiku)
• Standard analysis → scientist (Sonnet)
• Complex reasoning → scientist-high (Opus)

Session Management: Research state persists at .omc/research/{session-id}/ enabling resume after interruption.

---

Stopping Things

Just say:

• "stop"
• "cancel"
• "abort"

I'll intelligently determine what to stop based on context.

---

MCP Server Configuration

Extend Claude Code with additional tools via Model Context Protocol (MCP) servers.

/oh-my-claudecode:mcp-setup

Supported MCP Servers

Server	Description	API Key Required
Context7	Documentation and code context from popular libraries	No
Exa	Enhanced web search (replaces built-in websearch)	Yes
Filesystem	Extended file system access	No
GitHub	GitHub API for issues, PRs, repos	Yes (PAT)

Quick Setup

Run the setup command and follow the prompts:

/oh-my-claudecode:mcp-setup

Or configure manually in ~/.claude/settings.json:

{
  "mcpServers": {
    "context7": {
      "command": "npx",
      "args": ["-y", "@context7/mcp"]
    },
    "exa": {
      "command": "npx",
      "args": ["-y", "@anthropic/exa-mcp-server"],
      "env": {
        "EXA_API_KEY": "your-key-here"
      }
    }
  }
}

After configuration, restart Claude Code for changes to take effect.

---

What's Under the Hood

• 28 Specialized Agents - architect, researcher, explore, designer, writer, vision, critic, analyst, executor, planner, qa-tester, scientist (with tier variants)
• 31 Skills - orchestrate, ultrawork, ralph, planner, deepsearch, deepinit, git-master, frontend-ui-ux, learner, research, mcp-setup, and more
• MCP Server Support - Easy configuration of Context7, Exa, GitHub, and custom MCP servers
• Persistent Python REPL - True variable persistence for data analysis
• Research Workflow - Parallel scientist orchestration with /oh-my-claudecode:research command (new in 3.3.x)
• HUD Statusline - Real-time visualization of orchestration state
• Learned Skills - Extract reusable insights from sessions with /oh-my-claudecode:learner
• Memory System - Persistent context that survives compaction

---

HUD Statusline

The HUD displays real-time orchestration status in Claude Code's status bar:

[OMC] | 5h:0% wk:100%(1d6h) | ctx:45% | agents:Ae
todos:3/5 (working: Implementing feature)

Line 1: Core metrics

• Rate limits with reset times (e.g., wk:100%(1d6h) = resets in 1 day 6 hours)
• Context window usage
• Active agents (coded by type and model tier)

Line 2: Todo progress

• Completion ratio (3/5)
• Current task in progress

Run /oh-my-claudecode:hud setup to configure display options.

---

Coming from 2.x?

Good news: Your old commands still work!

/oh-my-claudecode:ralph "task"      →  Still works (or just say "ralph: task")
/oh-my-claudecode:ultrawork "task"  →  Still works (or just use "ulw" keyword)
/oh-my-claudecode:planner "task"    →  Still works (or just say "plan this")

The difference? You don't need them anymore. Everything auto-activates.

See the Migration Guide for details.

---

Documentation

• Full Reference - Complete documentation (800+ lines)
• Migration Guide - 2.x to 3.0 transition
• Architecture - Technical deep-dive
• Website - Online docs

---

Requirements

• Claude Code CLI
• One of:
  • Claude Max/Pro subscription (recommended for individuals)
  • Anthropic API key (for API-based usage)

---

Star History

---

License

MIT - see LICENSE

---

Inspired by:

oh-my-opencode • claude-hud • Superpowers • everything-claude-code

Zero learning curve. Maximum power.