| name | vibe |
| description | Vibe Coding - AI-driven development workflow with auto phase detection.
Analyzes user intent and automatically selects the right phase: Research, Plan, Implement, or Review.
Supports team mode for parallel execution with multiple agents.
Use when: "vibe", "분석", "계획", "구현", "리뷰", "analyze", "plan", "implement", "review", "build", "check code", "team", "팀", "solo", "싱글", "단독".
Single unified command replaces vibe-research, vibe-plan, vibe-implement, vibe-review.
|
| argument-hint | "<요청 내용>" or "team <요청 내용>" or "solo <요청 내용>" |
Vibe - AI Development Workflow
Single command, four phases. Analyzes your request and runs the right phase automatically.
Pipeline: Research (understand) -> Plan (decide) -> Implement (build) -> Review (verify)
No code is written until Phase 3. No implementation starts without an approved plan.
Phase Auto-Detection
Analyze the user's input and select ONE phase. If ambiguous, ask the user.
| Keywords | Phase | What Happens |
|---|
| 분석, 리서치, 조사, 구조, 의존성, 파악, research, analyze, investigate, explore | Research | Deep codebase analysis, no code changes |
| 계획, 플랜, 설계, plan, design, 세워줘, 전략 | Plan | Implementation plan with risk assessment |
| 구현, 만들어, 빌드, implement, build, code, 실행, 적용 | Implement | Code changes following approved plan |
| 리뷰, 검토, 코드리뷰, review, check, audit, 보안, PR | Review | Code quality, security, performance review |
Multi-Phase Execution
When the input requests multiple phases (e.g., "분석하고 계획까지", "전체 분석 후 구현 계획"), detect ALL requested phases and execute them sequentially.
Detection: Look for conjunctions ("하고", "후", "까지", "then", "and") connecting phase keywords. "전체" alone within a phase context means "all modes for that phase". "전체" with multiple phase keywords means "all requested phases".
Execution order: Always follow the pipeline order: Research -> Plan -> Implement -> Review. Skip phases not requested.
Output format for multi-phase:
## Phase 1/N: Research
[Complete research output following research.md template]
[Save to .vibe/NNN_topic/research.md]
---
## Phase 2/N: Plan
[Complete plan output following plan.md template]
[Save to .vibe/NNN_topic/plan.md]
[Reference the research.md just created above]
Each phase outputs its full template. All files go to the SAME topic folder.
Sub-option Auto-Detection
Within each phase, keywords activate specialized modes:
Research options:
| Keyword | Mode | Focus |
|---|
| 성능, 속도, 최적화, 메모리 | --deep | Performance profiling, complexity analysis |
| 패턴, 중복, 리팩토링, SOLID | --patterns | Anti-pattern detection, code smells |
| 의존성, 구조, 모듈, 순환 | --graph | Dependency graph, impact scope |
| 전체, 종합, 완전 | all modes | Complete analysis |
Plan options:
| Keyword | Mode | Focus |
|---|
| 피드백, 수정, 반영 | --feedback | Integrate inline MEMO comments |
| 리뷰, 평가, 검토 | --review | AI plan review and scoring |
| 리스크, 위험, 안전 | --risk-analysis | Detailed risk simulation |
Implement options:
| Keyword | Mode | Focus |
|---|
| 병렬, 동시, 빠르게 | --parallel | Parallel execution by dependency group |
| 안전, 롤백, 복원 | --rollback-on-fail | Auto-rollback on failure |
| 테스트, 시뮬레이션 | --dry-run | Preview changes without writing |
| 긴급, 핫픽스 | --fast | Minimal verification (use with caution) |
Review options:
| Keyword | Mode | Focus |
|---|
| 보안, 취약점, OWASP | --focus security | OWASP Top 10, auth/authz audit |
| 성능, 최적화 | --focus performance | Big O, re-renders, N+1, bundle size |
| 품질, 클린, SOLID | --focus quality | Code smells, duplication, patterns |
| PR, 머지, 풀리퀘스트 | --pr-ready | PR checklist and description |
| 엄격, strict | --strict | Zero tolerance for issues |
Team Mode
Automatically parallelizes work across multiple omc agents when complexity warrants it. Uses Claude Code's native team tools (TeamCreate, TaskCreate, Task, SendMessage, TeamDelete) with omc's specialized agent catalog.
Constraint: One team per session. The pipeline uses a single team with worker rotation between phases.
Mode Auto-Detection
After phase detection, evaluate complexity to decide Single vs Team mode:
| Phase | Team Mode Triggers | Single Mode |
|---|
| Research | sub-option 2+ activated, or 전체 keyword | sub-option 0-1 |
| Plan | multi-module scope (2+ directories), or --review + --risk-analysis both active | single module |
| Implement | 6+ changed files in plan.md, or team keyword | 5 or fewer files |
| Review | multi-focus (2+ focus areas), or --strict | single focus |
Team Mode Keywords
| Keywords | Effect |
|---|
| team, 팀 | Force team mode for all phases |
| solo, 싱글, 단독 | Force single mode (override auto-detection) |
N:agent-type (e.g., 3:executor) | Team mode + override Implement phase workers |
Keyword precedence: 병렬, 동시, parallel, concurrent remain as Implement sub-option (--parallel) triggers only. They do NOT force team mode. Only team/팀 and N:agent-type force team mode.
Team Declaration
When team mode activates, declare before starting:
[Phase]를 시작합니다. (Team Mode: N agents)
Topic: [topic]
Output: .vibe/NNN_topic/{phase_output}.md
Phase: [phase]
Mode: Team (auto-detected | user-specified)
Agents: [agent list with models]
Phase Agent Routing
Each phase spawns specialized workers. The lead (current session) orchestrates.
Research (Team):
| Task | subagent_type | Model | Focus |
|---|
| Codebase scan - file/function/type map | oh-my-claudecode:explore | haiku | Structure discovery |
| Performance profiling + pattern analysis | oh-my-claudecode:analyst | opus | --deep, --patterns |
| Dependency graph + risk analysis | oh-my-claudecode:architect | opus | --graph, risk |
Parallelism: All 3 simultaneous. Lead merges into research.md.
Plan (Team):
| Task | subagent_type | Model | Focus |
|---|
| Implementation strategy + file change plan | oh-my-claudecode:planner | opus | Core plan |
| Architecture review + alternatives | oh-my-claudecode:architect | opus | Review plan |
| Risk analysis + plan challenge | oh-my-claudecode:critic | opus | Challenge plan |
Execution: #1 first → #2, #3 parallel after plan draft exists. Lead integrates into plan.md.
Implement (Team):
| Task | subagent_type | Model | Focus |
|---|
| Group N file changes | oh-my-claudecode:executor | sonnet | Code changes |
| UI component work | oh-my-claudecode:designer | sonnet | Frontend |
| Test creation | oh-my-claudecode:test-engineer | sonnet | Tests |
Decomposition: Based on plan.md Execution Groups. Use TaskUpdate addBlockedBy for group ordering. Max 8 workers. Each worker runs tsc --noEmit after changes.
Review (Team):
| Task | subagent_type | Model | Focus |
|---|
| Code quality + architecture | oh-my-claudecode:code-reviewer | opus | Quality, SOLID |
| Security vulnerability scan | oh-my-claudecode:security-reviewer | sonnet | OWASP Top 10 |
| Test coverage + build verify | oh-my-claudecode:verifier | sonnet | Tests, build |
Parallelism: All 3 simultaneous. Lead merges into review.md.
User override: N:agent-type only affects Implement phase workers. For non-Implement phases, the override is ignored with a warning. Other phases use default routing.
Pipeline Orchestration
Single team, worker rotation between phases:
TeamCreate("vibe-NNN")
│
├─ Phase: Research (if team)
│ TaskCreate x N → Task(spawn workers) → Monitor → Collect → Shutdown workers
│ Lead writes research.md
│
├─ Phase: Plan (if team)
│ TaskCreate x N → Task(spawn workers) → Monitor → Collect → Shutdown workers
│ Lead writes plan.md
│ ⏸️ Approval gate (user approves plan.md)
│
├─ Phase: Implement (if team)
│ TaskCreate x N → Task(spawn workers) → Monitor → Validate → Shutdown workers
│ Completion report
│
├─ Phase: Review (if team)
│ TaskCreate x N → Task(spawn workers) → Monitor → Collect → Shutdown workers
│ Lead writes review.md
│
TeamDelete()
Worker spawn pattern:
{
"subagent_type": "oh-my-claudecode:{agent-type}",
"team_name": "vibe-NNN",
"name": "{phase}-{agent-type}[-N]",
"model": "{model}",
"prompt": "<vibe worker preamble + task instructions>"
}
Worker naming: research-explore, plan-planner, impl-executor-1, review-security.
Phase transition: Shutdown all current workers (SendMessage shutdown_request → wait for shutdown_response), then spawn next phase's workers in the same team.
Approval gate: Plan → Implement transition requires user approval (existing behavior). Auto-skip only if user explicitly requests ("전체 자동으로", "끝까지 알아서").
Worker Preamble
All workers receive this preamble (adapted per phase). All agents spawned as team members receive SendMessage capability from Claude Code's team infrastructure.
You are a VIBE TEAM WORKER in team "{team_name}". Your name is "{worker_name}".
You report to the team lead ("team-lead").
== VIBE CONTEXT ==
Project: {project_root}
Topic: .vibe/{NNN_topic}/
Phase: {current_phase}
Prior artifacts: {list of .vibe files to read}
== PHASE RULES ==
- Research: NEVER modify code. Analysis only.
- Plan: NEVER modify code. Planning only.
- Implement: ONLY within plan.md scope. Report out-of-scope to lead.
- Review: NEVER modify code (except --auto-fix). Review only.
== RESULT FORMAT ==
SendMessage to "team-lead" in markdown:
- Exact file_path:line_number references
- Severity classification (Critical/High/Medium/Low)
- Structured to match phase output template sections
== WORK PROTOCOL ==
1. TaskList → pick assigned pending task → TaskUpdate status "in_progress"
2. Execute task. Do NOT spawn sub-agents.
3. TaskUpdate status "completed"
4. SendMessage result to "team-lead"
5. Check TaskList for next task. If none, notify lead.
6. On shutdown_request → respond with shutdown_response.
Result Integration
Lead merges worker outputs into .vibe artifact sections:
| Phase | Worker → Sections |
|---|
| Research | explore → 1,2 / analyst → 5,6 / architect → 4,7 / Lead → 3,8-11 |
| Plan | planner → 0-4 / architect → 8 / critic → 9,11 / Lead → checklist |
| Review | code-reviewer → Quality,Summary / security → Security / verifier → Tests,PR / Lead → Overall Score, Action Items |
Handoff Between Phases
.vibe artifacts are the handoff. Each artifact includes a final section:
## Decisions & Rationale
- **Decided**: [key decisions]
- **Rejected**: [alternatives and why]
- **Risks**: [for next phase]
- **Remaining**: [items for next phase]
Lead reads prior artifacts before spawning next phase workers, including this section in their prompts.
| Transition | Handoff Document | Passed To Next Phase's Workers |
|---|
| Research → Plan | research.md (including Decisions & Rationale) | In planner/architect/critic prompts |
| Plan → Implement | plan.md (including Decisions & Rationale) | In executor/designer/test-engineer prompts |
| Implement → Review | git diff + plan.md | In reviewer/verifier prompts |
Error Handling
Worker failure:
- Research/Plan/Review: Retry failed subtask. Other results preserved.
- Implement: If --rollback-on-fail, rollback affected Group and retry.
Stuck agent detection:
- Task in_progress 5+ min without SendMessage → lead sends status check
- No response after 2 more min → reassign task to new worker
- Worker fails 2+ tasks → stop assigning to it
Phase failure:
- Shutdown remaining workers. Report to user. Suggest single-mode fallback.
- Do NOT TeamDelete on phase failure (team stays for retry or next phase).
Implement commit coordination:
- Workers report completion via SendMessage. Lead coordinates commit order to avoid conflicts.
.vibe Directory Convention
All artifacts live in .vibe/ at the project root. Each topic gets a numbered folder.
.vibe/
001_authentication/
research.md <- Phase 1 output
plan.md <- Phase 2 output
review.md <- Phase 4 output
002_payment/
research.md
plan.md
...
Indexing rules:
- Folder format:
NNN_topic_in_english_snake_case (3-digit zero-padded)
- Next index: count existing topic folders + 1
- Files:
research.md, plan.md, review.md inside each topic folder
- Tags in each file for cross-referencing:
#tags: auth, security, ...
NEXT_INDEX=$(printf "%03d" $(($(fd -t d -d 1 --exclude reviews --exclude checkpoints . .vibe 2>/dev/null | wc -l) + 1)))
TOPIC_DIR=".vibe/${NEXT_INDEX}_topic_name"
mkdir -p "$TOPIC_DIR"
Phase 1: Research
Goal: Deeply analyze the codebase. Produce a reusable research document. Write ZERO code.
Workflow
- Setup - Create
.vibe/NNN_topic/ folder, calculate index
- Declare scope - State the topic, activated modes, and output path
- Analyze - Read all relevant code with exact file:line references
- Write research.md - Extremely detailed, never summarize shallowly
Declare
Research를 시작합니다. 코드 변경 없이 분석만 진행합니다.
Topic: [주제]
Output: .vibe/NNN_topic/research.md
Modes: [기본 | deep | patterns | graph]
Detected keywords: [...]
research.md Structure
# Research: [주제]
**Date**: YYYY-MM-DD HH:MM
**Index**: NNN
**Status**: research-only
**Modes**: [activated modes]
**Tags**: #tag1 #tag2
---
## Related Research
- [NNN_other_topic](../NNN_other_topic/research.md) - description
## 1. File/Function/Type Map
| File Path | Role | Complexity | Change Risk |
|-----------|------|-----------|-------------|
| src/path:lines | description | Low/Med/High | Low/Med/High |
## 2. Current Flow
[mermaid sequence diagram: trigger -> process -> output]
## 3. Data Flow
- Input sources, validation, transformations
- State management, storage, external APIs
## 4. Dependencies & Side Effects
[mermaid dependency graph]
## 5. Pattern Analysis (--patterns)
| Pattern | Location | Frequency | Action |
|---------|----------|-----------|--------|
[Anti-patterns, code smells, duplication]
## 6. Performance Analysis (--deep)
| Function | Time | Space | Optimizable |
|----------|------|-------|------------|
[Big O, bottlenecks, N+1, memory leaks]
## 7. Risk & Impact Scope
| Risk | Severity | Scope | Action |
|------|----------|-------|--------|
[Security, breaking changes, blast radius]
## 8. Unknowns & Questions
- [ ] Items requiring user confirmation
## 9. Metrics
[File count, line count, test coverage, complexity stats]
## 10. Summary & Next Steps
- Key findings (numbered)
- Immediate actions needed
- **Next: `/vibe "계획 세워줘"` to create implementation plan**
## Decisions & Rationale
- **Decided**: [key decisions made during analysis]
- **Rejected**: [alternatives considered and why]
- **Risks**: [identified risks for planning phase]
- **Remaining**: [items left for plan phase to address]
Analysis Tools
ast-grep (sg): Structural code pattern search
rg (ripgrep): Fast text search
fd: File discovery
tokei: Code statistics
Phase 2: Plan
Goal: Create a detailed implementation plan. Write ZERO code. Developer approves before any implementation.
Workflow
- Find research - Locate latest
.vibe/NNN_topic/research.md (or specified with --research NNN_topic)
- If NO research.md exists: stop and tell the user
"research.md가 없습니다. 먼저 /vibe "주제 분석" 을 실행하세요." Do NOT proceed without research.
- If research exists but is outdated: warn and ask whether to proceed or re-research
- Read & confirm - Fully read research, state summary
- Write plan.md - Detailed plan with MEMO spaces and AI review
- Get approval - Plan stays DRAFT until explicitly approved
plan.md Structure
# Plan: [주제]
**Based on**: .vibe/NNN_topic/research.md
**Date**: YYYY-MM-DD HH:MM
**Status**: DRAFT (미승인 - 구현 금지)
**Approval**: [ ] 미승인
**Risk Level**: [Low/Medium/High/Critical]
**Estimated Time**: [N hours/days]
---
## 0. Goals & Non-Goals
### Goals
- [ ] [Specific, measurable goal]
### Non-Goals
- [Explicitly excluded items]
<!-- MEMO: -->
## 1. File Changes
### New Files
| Path | Purpose | Est. Lines | Priority |
### Modified Files
| Path | Change Type | Impact | Risk |
### Deleted Files
| Path | Reason | Dependencies Checked |
<!-- MEMO: -->
## 2. Per-File Change Details
### path/to/file.ts
**Before:** [current code snippet]
**After:** [planned code change]
**Reason:** [why this change]
<!-- MEMO: -->
## 3. Type/Interface Changes
[Breaking vs backward-compatible changes with code examples]
<!-- MEMO: -->
## 4. Implementation Strategy
[mermaid graph of phase order]
### Phase N: [Name] (time estimate)
- [ ] Task items
<!-- MEMO: -->
## 5. Migration & Compatibility
[Strategy, compatibility matrix, breaking change notices]
<!-- MEMO: -->
## 6. Test Strategy
| Type | Current | Target | New Tests |
[Test scenarios: happy path, errors, edge cases]
<!-- MEMO: -->
## 7. Rollback Strategy
[Triggers, procedure, complexity assessment]
<!-- MEMO: -->
## 8. Alternatives & Tradeoffs
[Considered alternatives with pros/cons comparison table]
<!-- MEMO: -->
## 9. Risk Analysis
| Risk | Probability | Impact | Level | Mitigation |
[Risk score calculation, contingency plans]
<!-- MEMO: -->
## 10. Key Decision Questions
- [ ] **Q1**: [Decision needed]
<!-- MEMO: -->
## 11. AI Review Report
[Completeness score, missing elements, improvement suggestions, alternative approaches]
## 12. Decisions & Rationale
- **Decided**: [key decisions made during planning]
- **Rejected**: [alternatives considered and why]
- **Risks**: [identified risks for implementation]
- **Remaining**: [items left for implement phase to address]
<!-- MEMO: -->
## Approval Checklist
- [ ] Research up to date
- [ ] Goals/non-goals clear
- [ ] File paths confirmed
- [ ] Test strategy confirmed
- [ ] Rollback strategy confirmed
- [ ] Risk level acceptable
- [ ] AI review score > 80
- [ ] **Developer final approval**: [ ]
**Next: After approval, run `/vibe "구현해"` to start implementation**
--feedback Mode
Extracts all <!-- MEMO: comments --> from existing plan.md and updates the plan accordingly.
--review Mode
AI evaluates the plan: completeness score, missing considerations, risk assessment.
Phase 3: Implement
Goal: Execute the approved plan mechanically. Stay within plan scope. Verify continuously.
Pre-Implementation Gate
Automatically checks plan approval status:
LATEST_TOPIC_DIR=$(fd -t d -d 1 '^[0-9]' .vibe 2>/dev/null | sort -r | head -1)
rg "Developer final approval.*\[x\]" "${LATEST_TOPIC_DIR}/plan.md"
If not approved: block implementation, show missing checklist items.
Workflow
- Verify approval - Check plan.md approval status
- Create branch -
git checkout -b vibe-impl-YYYYMMDD-HHMMSS
- Analyze dependencies - Build execution groups from plan
- Implement by group - Sequential or parallel, with checkpoints
- Validate continuously - Type check + lint + test after each file
- Report completion - Summary with metrics and next steps
Declare
Implementation을 시작합니다.
Plan: .vibe/NNN_topic/plan.md
Research: .vibe/NNN_topic/research.md (auto-referenced)
Files: N개
Risk Level: [level]
Mode: [Sequential | Parallel]
Safety: [typecheck: on | rollback: on/off | watch: on/off]
Execution Groups
Classify files by dependency order:
Group 1 (parallel OK): types, constants, config
Group 2 (after G1): utilities, helpers
Group 3 (after G2): core logic, API routes
Group 4 (after G3): components, pages
Group 5 (after G4): tests, docs
Continuous Validation
After each file change:
tsc --noEmit (type check)
- Lint check
- Related tests
After each group:
- Full type check
- Full test suite
- Create checkpoint:
git add -A && git commit -m "checkpoint: group N complete"
Out-of-Scope Changes
When discovering needed changes outside plan scope:
Plan 범위 외 변경 감지
Location: [file:line]
Issue: [description]
Severity: [Low/Medium/High/Critical]
Options:
A) Emergency patch (implement + update plan)
B) Add TODO comment (defer)
C) Stop implementation (re-plan)
D) Separate hotfix branch
Only option A for Critical security issues. Otherwise, ask the user.
Completion Report
Implementation 완료!
Changes: N files, +X/-Y lines
Validation: typecheck [pass/fail] | lint [pass/fail] | tests [pass/fail]
Coverage: X% (+N%)
Out-of-scope items: [list if any]
Branch: vibe-impl-YYYYMMDD-HHMMSS
**Next: `/vibe "코드 리뷰"` to review changes**
Rollback Protocol (--rollback-on-fail)
Triggers: build failure, >5 type errors, >10% test failure rate, security vulnerability.
Action: revert to last checkpoint, log failure reason.
Phase 4: Review
Goal: Comprehensive code review across security, performance, quality, and test coverage.
Workflow
- Collect changes -
git diff main...HEAD, load .vibe/ context
- Declare scope - State review range, focus areas, activated modes
- Multi-dimensional analysis - Security, performance, quality, tests
- Generate report - review.md with actionable items and PR readiness
- Auto-fix (if --auto-fix) - Apply fixable issues automatically
Declare
Code Review를 시작합니다.
Branch: [current branch]
Changed files: N개
Lines: +X/-Y
Focus: [security, performance, quality, testing, all]
Mode: [standard | strict | pr-ready]
review.md Structure
# Code Review Report
**Date**: YYYY-MM-DD HH:MM
**Branch**: [branch name]
**Overall Score**: XX/100
---
## Executive Summary
### Strengths
- [Good patterns found]
### Issues
- [Critical/High/Medium items count]
## Security Analysis
### [Critical/High/Medium/Low] Issues
For each issue:
- **File**: path:line
- **Issue**: description
- **Fix**: code example (before/after)
## Performance Analysis
### Optimization Opportunities
For each:
- **Location**: path:line
- **Impact**: High/Medium/Low
- **Fix**: code example
## Code Quality
- Maintainability score
- SOLID compliance check
- Code smells and duplication
- Design pattern suggestions
## Test Coverage
| Type | Coverage | Target | Gap |
[Uncovered critical paths listed]
## PR Readiness Checklist
### Must Fix (blocks PR)
- [ ] [Critical items]
### Should Fix
- [ ] [Important items]
### Nice to Have
- [ ] [Improvements]
## Action Items (Prioritized)
1. **P0 Security**: [immediate fixes]
2. **P1 Performance**: [this sprint]
3. **P2 Quality**: [tech debt backlog]
**Next: Fix issues, then create PR or run `/vibe "리뷰"` again to verify**
--pr-ready Mode
Generates a complete PR description with summary, changes, metrics, breaking changes, and test instructions.
--strict Mode
Zero tolerance: coverage > 90%, no security issues, zero lint errors.
--auto-fix Mode
Automatically applies fixable issues (lint, formatting, simple patterns). Logs all changes.
Critical Rules
All Phases
- Reference code with exact
file_path:line_number format
- Surface unknowns explicitly - never hide uncertainty
- Use mermaid diagrams for flows and dependencies
- Use Rust CLI tools:
rg, fd, ast-grep, bat, eza, tokei
- Always state the detected phase and activated options at the start
- Always indicate the next step/command at the end
Phase-Specific Prohibitions
| Phase | NEVER Do |
|---|
| Research | Write or modify any code files |
| Plan | Write code before explicit developer approval |
| Implement | Change anything outside the approved plan scope without asking |
| Review | Ignore Critical security issues or approve without tests |
Quality Gates
| Phase | Gate |
|---|
| Research | All references must be verifiable file:line paths |
| Plan | AI review score > 80, all key questions answered |
| Implement | Type check pass, tests pass, zero security vulnerabilities |
| Review | All Critical/High issues documented with fix examples |