omc (oh-my-claudecode) — Claude Code Multi-Agent Orchestration

When to use this skill

• You want Teams-first multi-agent orchestration inside Claude Code
• You need 29+ specialized agents with smart model routing (Haiku → Opus)
• Complex tasks that benefit from parallel agent execution with verify/fix loops
• Any Claude Code workflow that needs persistent, guaranteed-completion execution
• You want skill layers: Guarantee (ralph) → Enhancement (ultrawork) → Execution
• You need 3-model synthesis with Codex + Gemini + Claude (CCG mode)

---

1. Installation (3 Steps)

Step 1: Install 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

Step 3: Build something

autopilot: build a REST API for managing tasks

npm alternative: npm install -g oh-my-claude-sisyphus

Enable native teams (~/.claude/settings.json):

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

---

2. Orchestration Modes

Mode	What it is	Use For
Team (canonical)	5-stage pipeline: plan → prd → exec → verify → fix	Coordinated agents on shared task list
Autopilot	Full autonomous 5-stage pipeline	End-to-end feature work (idea → code)
Ultrawork	Maximum parallelism burst	Fast fixes/refactors across codebase
Ralph	Persistent loop: execute → verify → fix	Tasks that must complete fully, no partials
Pipeline	Sequential staged processing	Strict ordering / multi-step transforms
CCG	3-model synthesis (Codex + Gemini + Claude)	Mixed backend + UI work
omc team (CLI)	tmux workers with Codex/Gemini/Claude	Cross-model code review, security, design

Skill Layer Composition

┌──────────────────────────────────────────┐
│ GUARANTEE LAYER (optional)               │
│ ralph — won't stop until verified done   │
└──────────────────────────────────────────┘
                    │
                    ▼
┌──────────────────────────────────────────┐
│ ENHANCEMENT LAYER (0–N skills)           │
│ ultrawork (parallel) | git-master (VCS)  │
└──────────────────────────────────────────┘
                    │
                    ▼
┌──────────────────────────────────────────┐
│ EXECUTION LAYER (primary skill)          │
│ default (build) | planner | orchestrate  │
└──────────────────────────────────────────┘

---

3. Magic Keywords (Auto-Detection)

Keyword	Effect
autopilot, build me, I want a	Full autonomous pipeline
ralph, don't stop, must complete	Persistent loop until verified
ulw, ultrawork	Maximum parallelism
ccg, claude-codex-gemini	3-model synthesis
ralplan	Consensus-based iterative planning
deep interview, ouroboros	Socratic requirements clarification
deepsearch	Codebase-focused deep search
deepanalyze, deep-analyze	Deep analysis mode
ultrathink	Deep reasoning mode
tdd, test first	TDD workflow
deslop, anti-slop	Clean AI expression patterns
cancelomc, stopomc	Cancel active execution mode
plan	Planning interview

---

4. Team Mode (Canonical)

/oh-my-claudecode:team 3:executor "fix all TypeScript errors"

5-stage pipeline:

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

• team-plan: Planning agent creates task list
• team-prd: Architect designs solution
• team-exec: N parallel executor agents implement
• team-verify: Verifier confirms completion
• team-fix: If verification fails, fix loop repeats

---

5. Agent System (29+ Specialized Agents)

Build / Analysis Lane: explore (haiku) · analyst (opus) · planner (opus) · architect (opus) · debugger (sonnet) · executor (sonnet) · verifier (sonnet) · tracer (sonnet)

Review Lane: code-reviewer (opus) · security-reviewer (sonnet) · critic (opus) · code-simplifier (opus)

Domain Specialists: test-engineer · designer · writer · qa-tester · scientist · git-master · document-specialist

Invoke via: /oh-my-claudecode:<agent-name> or delegated through Task tool.

See references/agents-catalog.md for full profiles.

---

6. CLI Tools

omc team (tmux Workers)

omc team 2:codex "review auth module for security"
omc team 2:gemini "redesign UI components"
omc team status <session>
omc team shutdown <session>

omc ask (Provider Advisor)

omc ask claude "review this migration plan"
omc ask codex --prompt "identify architecture risks"
omc ask gemini --prompt "propose UI polish ideas"

omc wait (Rate Limit Auto-Resume)

omc wait              # Check status
omc wait --start      # Enable auto-resume daemon
omc wait --stop       # Disable daemon

Notifications

omc config-stop-callback telegram --enable --token <bot> --chat <id>
omc config-stop-callback discord --enable --webhook <url>
omc config-stop-callback slack --enable --webhook <url>

---

7. Updating

# Sync latest version
/plugin marketplace update omc

# Re-run setup
/oh-my-claudecode:omc-setup

# If issues after update
/oh-my-claudecode:omc-doctor

---

8. Optional: Multi-AI Orchestration

Provider	Install	What it enables
Gemini CLI	npm install -g @google/gemini-cli	Design review, UI consistency (1M token context)
Codex CLI	npm install -g @openai/codex	Architecture validation, code review cross-check

---

Instructions

1. Detect intent — match user request to the best orchestration mode (Team, Autopilot, Ralph, Ultrawork, CCG)
2. Provide install path — if OMC is not yet installed, give the 3-step install sequence
3. Show the invocation — give the exact command or magic keyword for the chosen mode
4. Set expectations — explain what the mode does (pipeline stages, loop behavior, parallelism)
5. Surface relevant CLI — mention omc team, omc ask, or omc wait if they apply
6. Link to references — point to the appropriate references/ file for deep-dive details

---

Examples

Example 1: Multi-agent team task

User: "여러 에이전트가 병렬로 TypeScript 에러 전부 고쳐줘" Response: Use Team mode — /oh-my-claudecode:team 3:executor "fix all TypeScript errors". Runs 3 parallel executors through plan→prd→exec→verify→fix pipeline.

Example 2: Guaranteed completion

User: "테스트 전부 통과할 때까지 멈추지 마" Response: Use Ralph mode — ralph: fix all failing tests. Persistent execute→verify→fix loop until all tests pass. Won't stop until verified.

Example 3: New installation

User: "OMC 설치하고 싶어" Response: 3-step install: /plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode → /plugin install oh-my-claudecode → /oh-my-claudecode:omc-setup. Enable native teams in ~/.claude/settings.json.

---

Best practices

• Default to Team mode for multi-agent work — swarm/ultrapilot are legacy and route to Team
• Enable CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in ~/.claude/settings.json before using Team mode
• Use ralph for critical tasks — it composes automatically with ultrawork for parallel+persistent execution
• Use CCG for mixed workloads — backend execution (Codex) + large-context analysis (Gemini) + orchestration (Claude)
• Re-run /oh-my-claudecode:omc-setup after every plugin update to refresh hooks and CLAUDE.md
• npm package name differs — it's oh-my-claude-sisyphus, not oh-my-claudecode
• Cancel safely — use cancelomc keyword or /oh-my-claudecode:cancel; state is preserved for resume

---

Quick Reference

Action	Command
Install	/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
Setup	/oh-my-claudecode:omc-setup
Team mode	/oh-my-claudecode:team N:executor "task"
Autopilot	autopilot: <task>
Ralph loop	ralph: <task>
Ultrawork	ulw <task>
CCG synthesis	ccg: <task>
Deep interview	deep interview: <topic>
Cancel	cancelomc
Update	/plugin marketplace update omc && /oh-my-claudecode:omc-setup
Debug	/oh-my-claudecode:omc-doctor

→ Full documentation · GitHub

---

References

File	Contents
references/agents-catalog.md	All 29+ agent profiles, tiers, delegation rules
references/modes-reference.md	Mode deep-dives, skill layer composition
references/hooks-reference.md	Hook events, injection patterns, kill switches
references/cli-reference.md	Full CLI syntax: omc team/ask/wait/config