원클릭으로
agent-catalog
Agent directory — lists all omg agents with their roles, models, and when to use each
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
Agent directory — lists all omg agents with their roles, models, and when to use each
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | agent-catalog |
| description | Agent directory — lists all omg agents with their roles, models, and when to use each |
| tags | ["reference","delegation"] |
Use this catalog to decide which agent to spawn for a task. All agents are invoked via the task tool:
task(agent_type="omg:<name>", prompt="<task>", model="<model>", mode="background|sync")
# With effort control (Copilot CLI --effort flag):
copilot -p "<task>" --agent omg:<name> --effort <low|medium|high|xhigh> --yolo
READ-ONLY agents should be spawned with --excluded-tools to physically prevent file modifications:
# When spawning via copilot CLI (automation/CI):
copilot -p "..." --agent omg:architect --excluded-tools edit,create --yolo
# The task() tool does not support --excluded-tools directly,
# but the agent prompt enforces read-only behavior.
# For maximum security in CI, use copilot CLI with --excluded-tools.
| Agent | task() name | Purpose | Mode |
|---|---|---|---|
| autopilot | omg:autopilot | Full lifecycle: idea → working verified code. Delegates to specialists. | sync |
| ralph | omg:ralph | Persistence loop — keeps iterating until all acceptance criteria pass. | sync |
| team | omg:team | Decomposes task, fires N parallel workers, verifies combined result. | sync |
| ralplan | omg:ralplan | Consensus planning: planner → architect → critic loop until agreement. | sync |
| ultrawork | omg:ultrawork | Parallel fire-and-forget — launches independent tasks simultaneously. | sync |
| research-to-pr | omg:research-to-pr | Deep investigation → /delegate → cloud agent creates PR. Copilot-exclusive. | sync |
| sciomc | omg:sciomc | Staged parallel research with tiered scientist agents. | sync |
| self-improve | omg:self-improve | Autonomous optimization: benchmark, tournament-select best approach. | sync |
| deep-dive | omg:deep-dive | Two-stage: trace root cause → crystallize into actionable spec. | sync |
| Agent | task() name | Purpose | Mode |
|---|---|---|---|
| architect | omg:architect | Code analysis, root-cause diagnosis, architectural recommendations. READ-ONLY. | sync |
| critic | omg:critic | Final quality gate — multi-perspective review with severity ratings. READ-ONLY. | sync |
| analyst | omg:analyst | Requirements analysis — catches gaps, edge cases, undefined guardrails. READ-ONLY. | sync |
| code-reviewer | omg:code-reviewer | Severity-rated code review — spec compliance, security, SOLID, logic. READ-ONLY. | sync |
| security-reviewer | omg:security-reviewer | OWASP Top 10, secrets scanning, dependency audits. READ-ONLY. | sync |
| code-simplifier | omg:code-simplifier | Simplifies code for clarity without changing behavior. | sync |
| Agent | task() name | Purpose | Mode |
|---|---|---|---|
| executor | omg:executor | Implements code changes — smallest viable diff, verified. | background |
| debugger | omg:debugger | Root-cause analysis, minimal fixes, build error resolution. | sync |
| verifier | omg:verifier | Evidence-based completion checks — fresh test output, PASS/FAIL. | sync |
| test-engineer | omg:test-engineer | Test strategy, TDD, flaky test hardening. | background |
| designer | omg:designer | UI/UX — stunning, production-grade interfaces. | background |
| git-master | omg:git-master | Atomic commits, style-matched messages, safe rebasing. | sync |
| qa-tester | omg:qa-tester | Interactive CLI testing via tmux sessions. | sync |
| scientist | omg:scientist | Data analysis with statistical rigor. READ-ONLY. | sync |
| tracer | omg:tracer | Evidence-driven causal tracing with competing hypotheses. READ-ONLY. | sync |
| document-specialist | omg:document-specialist | External docs, API references, version compatibility. READ-ONLY. | background |
| Agent | task() name | Purpose | Mode |
|---|---|---|---|
| explore | omg:explore | Fast codebase search — finds files, patterns, relationships. READ-ONLY. | background |
| writer | omg:writer | Technical documentation with verified code examples. | background |
Full lifecycle from idea to code?
→ omg:autopilot (sonnet, sync)
Must complete with proof?
→ omg:ralph (sonnet, sync)
Multiple independent subtasks?
→ omg:team (sonnet, sync)
Need consensus plan?
→ omg:ralplan (opus, sync)
Investigate → auto PR?
→ omg:research-to-pr (sonnet, sync)
Need to find files/patterns?
→ omg:explore (haiku, background)
Need to understand architecture?
→ omg:architect (opus, sync)
Need to implement code?
→ omg:executor (sonnet, background)
Need to fix a bug?
→ omg:debugger (sonnet, sync)
Need to verify work is done?
→ omg:verifier (sonnet, sync)
Need a code review?
→ omg:code-reviewer (opus, sync)
Need a security audit?
→ omg:security-reviewer (opus, sync)
Need to plan before coding?
→ omg:analyst (opus, sync) then omg:planner (opus, sync)
Need a quality gate on a plan?
→ omg:critic (opus, sync)
Need tests written?
→ omg:test-engineer (sonnet, background)
Need external docs?
→ omg:document-specialist (sonnet, background)
Need clean git history?
→ omg:git-master (sonnet, sync)
# Fast codebase search
task(agent_type="omg:explore", prompt="find all auth-related files", model="claude-haiku-4.5", mode="background")
# Deep architecture review
task(agent_type="omg:architect", prompt="analyze the auth module design", model="claude-opus-4.6", mode="sync")
# Implementation
task(agent_type="omg:executor", prompt="add input validation to createUser()", model="claude-sonnet-4.6", mode="background")
# Verification
task(agent_type="omg:verifier", prompt="verify createUser validation works", model="claude-sonnet-4.6", mode="sync")
agent catalog, list agents, which agent
Detect and clean AI-typical writing patterns in documents — em dashes, filler phrases, over-formality
Route questions to Claude, Codex, or Gemini for multi-perspective answers
Claude-Codex-Gemini tri-model orchestration — decompose, query, synthesize
Health check — verifies omg plugin is correctly installed and working
Quick reference — what omg agents and skills are available and how to use them
About omg — version, architecture, capabilities, and credits