omc-tips

OMC Tips — oh-my-claudecode Quick Reference

oh-my-claudecode (OMC) is a multi-agent orchestration plugin for Claude Code. It turns Claude Code into a platform of 19 specialized agents, 37 skills, and 20 lifecycle hooks — all driven by natural-language magic keywords.

Architecture at a Glance

autopilot   (5-stage pipeline: idea → working code)
  └── ralph   (persistence loop — keeps going until verified)
       └── ultrawork   (parallel agent execution engine)
            └── agents   (19 specialists across 3 model tiers)

Core flow: User Input → Hooks (event detection) → Skills (behavior injection) → Agents (task execution) → State (progress tracking)

Planning pipeline: deep-interview → ralplan/plan → autopilot Investigation pipeline: deep-dive → trace + deep-interview → plan → autopilot

---

Mode Selection — Decision Tree

Is the task vague/unclear?
  ├─ Yes → deep-interview (then autopilot)
  └─ No → Complex multi-step project?
            ├─ Yes → Multiple independent components?
            │         ├─ Yes → team (coordinated pipeline)
            │         └─ No  → autopilot (single-lead autonomous)
            └─ No → Is it a bug?
                      ├─ Yes → Cause unclear?
                      │         ├─ Yes → deep-dive or trace
                      │         └─ No  → ralph (persist until fixed)
                      └─ No → Parallelizable?
                                ├─ Yes → ulw (ultrawork)
                                └─ No  → Just describe the task normally

Mode	What It Does	When To Use
autopilot	5-stage pipeline (Expansion→Planning→Execution→QA→Validation)	Single coherent deliverable
team	3-5 parallel agents with file ownership	3+ independent subtasks
ralph	Verification loop — repeats until done	Completion guarantee needed
ultrawork (ulw)	Maximum parallelism + manual oversight	Speed with control
ccg	Claude+Codex+Gemini tri-model synthesis	Multi-perspective validation
ralplan	Planner→Architect→Critic consensus loop	Unclear requirements
deep-interview	Socratic requirements clarification	Vague starting point

Invalid combos: Two standalone modes together (e.g., autopilot team) — the second keyword is silently ignored. eco is a modifier only — pair it with autopilot/ralph/ultrawork.

---

Magic Keywords — Natural Language Triggers

Include these anywhere in a prompt to activate the corresponding mode:

Keyword	Activates	Example
autopilot, build me	Full autonomous pipeline	"autopilot: build a REST API"
ralph, must complete	Persistence loop	"ralph: fix the auth bug completely"
ulw, ultrawork	Parallel execution	"ulw refactor the API layer"
team	Coordinated agents	/team 3:executor fix errors
ccg	Tri-model synthesis	"ccg: GraphQL vs REST?"
ralplan	Consensus planning	"ralplan this feature"
deep-interview	Socratic questioning	"deep-interview: build a dashboard"
deslop	AI slop cleaner	"deslop the recent changes"
deepsearch	Codebase search	"deepsearch: where is auth?"
ultrathink	Deep reasoning	"ultrathink: why is this deadlocking?"
cancelomc	Stop active mode	"cancelomc"

Priority: cancel (exclusive) → ralph → autopilot → ultrawork → others

---

Agents — 19 Specialists in 3 Tiers

Tier	Model	Agents
Opus (deep)	opus	analyst, planner, architect, critic, code-reviewer, code-simplifier
Sonnet (standard)	sonnet	executor, debugger, verifier, tracer, test-engineer, designer, qa-tester, scientist, document-specialist, git-master, security-reviewer
Haiku (fast)	haiku	explore, writer

Typical sequence: explore → analyst → planner → critic → executor → verifier

Some agents have tier variants (e.g., architect-low, executor-high, security-reviewer-low). Override model via model parameter in Agent tool calls: model: "opus" on executor for complex tasks. See references/detailed-reference.md for full tier variant list.

For detailed agent capabilities, tool assignments, and lane descriptions, consult references/detailed-reference.md.

---

Key Skills Quick Reference

Execution & Orchestration

Skill	Command	Notes
autopilot	/oh-my-claudecode:autopilot	5-stage autonomous pipeline
ralph	/oh-my-claudecode:ralph	Loop until verified. Includes ultrawork.
ultrawork	/oh-my-claudecode:ultrawork	Max parallel agents
team	/oh-my-claudecode:team N:type "task"	N agents, types: executor/debugger/designer/codex/gemini
ultraqa	/oh-my-claudecode:ultraqa	QA cycle: test→verify→fix→repeat
cancel	/oh-my-claudecode:cancel	Stop any active mode

Planning & Analysis

Skill	Command
plan	/oh-my-claudecode:plan
ralplan	/oh-my-claudecode:ralplan
deep-interview	/oh-my-claudecode:deep-interview
deep-dive	/oh-my-claudecode:deep-dive
trace	/oh-my-claudecode:trace

Multi-Model & Research

Skill	Command
ccg	/oh-my-claudecode:ccg
ask	/oh-my-claudecode:ask — route to claude/codex/gemini
omc-teams	/oh-my-claudecode:omc-teams — tmux CLI workers
sciomc	/oh-my-claudecode:sciomc — parallel scientist agents
external-context	/oh-my-claudecode:external-context — web doc lookup

Code Quality & Knowledge

Skill	Command
ai-slop-cleaner	/oh-my-claudecode:ai-slop-cleaner
visual-verdict	/oh-my-claudecode:visual-verdict
deepinit	/oh-my-claudecode:deepinit — generate AGENTS.md across codebase
verify	/oh-my-claudecode:verify
self-improve	/oh-my-claudecode:self-improve

Setup & Admin

Skill	Command
omc-setup	/oh-my-claudecode:omc-setup
omc-doctor	/oh-my-claudecode:omc-doctor
hud	/oh-my-claudecode:hud — statusline: minimal/focused/full
mcp-setup	/oh-my-claudecode:mcp-setup

---

Practical Examples

Scenario	Command
Vague idea → code	deep-interview "real-time dashboard for metrics"
Fix all errors	/team 3:executor "fix all TS errors in src/"
Large refactoring	ulw refactor user API into service/controller/repo
Mystery bug	deep-dive "intermittent 403 on /api/orders"
Must-complete task	ralph: migrate raw SQL to repository pattern
Multi-AI opinion	ccg: should we use GraphQL or REST?
Security review	review payment module for security issues
Clean AI slop	deslop the recently modified files
Feature planning	ralplan: implement OAuth2 with PKCE
Map new codebase	/oh-my-claudecode:deepinit

---

State & Memory

Path	Purpose	Survives Compaction?
.omc/state/	Per-mode state	Yes
.omc/notepad.md	Working memory	Yes
.omc/project-memory.json	Project knowledge	Yes (permanent)
.omc/plans/	Execution plans	Yes
.omc/skills/	Project-level custom skills	N/A

Memory tags: <remember> (7 days), <remember priority> (permanent)

---

Environment Variables

Variable	Purpose	Default
DISABLE_OMC	Disable all hooks	—
OMC_SKIP_HOOKS	Skip specific hooks (comma-separated)	—
OMC_STATE_DIR	Centralized state directory	.omc/state/
OMC_PARALLEL_EXECUTION	Toggle parallel execution	true
OMC_LSP_TIMEOUT_MS	LSP request timeout	15000

---

CLI Helpers

omc ask claude/codex/gemini "prompt"  # Route to specific AI
omc team N:type "task"                # Start team job
omc team status                       # Inspect running jobs
omc team shutdown <name> --force      # Stop team job
omc hud                               # Render statusline
omc wait --start                      # Auto-resume on rate limit
omc session search "query" --since 7d # Search past sessions

---

Additional Resources

For in-depth details on architecture, hooks, MCP tools, verification tiers, team pipelines, and cost optimization strategies, consult:

• references/detailed-reference.md — Complete architecture, hooks system, MCP tools, verification protocol, team mode details, and advanced strategies