@su-record/vibe 2.5.9 → 2.5.11

package/CLAUDE.md

@@ -1,348 +1,448 @@
-# VIBE
-
-SPEC-driven AI Coding Framework (Claude Code Exclusive)
-
-## Response Language
-
-**IMPORTANT: Always respond in English unless the user explicitly requests otherwise.**
-
-## Response Language
-
-**IMPORTANT: Always respond in Korean (한국어) unless the user explicitly requests otherwise.**
-
-## Code Quality Standards (Mandatory)
-
-Follow these standards when writing code. See `~/.claude/vibe/rules/` (global) for detailed rules.
-
-### Core Principles
-- **Modify only requested scope** - Don't touch unrelated code
-- **Preserve existing style** - Follow project conventions
-- **Keep working code** - No unnecessary refactoring
-- **Respect user interrupts** - If user interrupts (Ctrl+C/Escape) and sends a new message, the previous task is CANCELLED. Do NOT resume or continue interrupted work. Respond ONLY to the new message.
-
-### Code Complexity Limits
-| Metric | Limit |
-|--------|-------|
-| Function length | ≤30 lines (recommended), ≤50 lines (allowed) |
-| Nesting depth | ≤3 levels |
-| Parameters | ≤5 |
-| Cyclomatic complexity | ≤10 |
-
-### TypeScript Rules
-- No `any` type → Use `unknown` + type guards
-- No `as any` casting → Define proper interfaces
-- No `@ts-ignore` → Fix type issues at root
-- Explicit return types on all functions
-
-### TypeScript Rules
-- No `any` type → use `unknown` + type guards
-- No `as any` casting → define proper interfaces
-- No `@ts-ignore` → fix type issues at root
-- Explicit return types on all functions
-
-### Error Handling Required
-- try-catch or error state required
-- Loading state handling
-- User-friendly error messages
-
-### Forbidden Patterns
-- No console.log in commits (remove after debugging)
-- No hardcoded strings/numbers → Extract to constants
-- No commented-out code in commits
-- No incomplete code without TODO
-
-## Workflow
-
-```
-/vibe.spec → (auto) SPEC review → /vibe.run → (auto) code review → (auto) fix → ✅ Done
-```
-
-**Automated Flow:**
-1. `/vibe.spec` - Write SPEC + **(auto)** Gemini review → Auto-apply
-2. `/vibe.run` - Implementation + Gemini review
-3. **(auto)** 13+ agent parallel review
-4. **(auto)** P1/P2 issue auto-fix
-
-## Plan Mode vs VIBE (Workflow Selection)
-
-**Offer choice to user on development requests:**
-
-| Task Size | Recommended |
-|-----------|-------------|
-| Simple changes (1-2 files) | Plan Mode |
-| Complex features (3+ files, research/verification needed) | `/vibe.spec` |
-
-| Item | Plan Mode | VIBE |
-|------|-----------|------|
-| Storage location | `~/.claude/plans/` (global) | `.claude/vibe/specs/` (project) |
-| Document format | Free form | PTCF structure (AI-optimized) |
-| Research | None | 4 parallel agents |
-| Verification | None | `/vibe.verify` against SPEC |
-| History | Not trackable | Git version control |
-
-**Rules:**
-- After `/vibe.analyze` or `/vibe.review` with dev/modify request → **Ask user for workflow choice**
-- User chooses VIBE → Wait for `/vibe.spec`
-- User chooses Plan Mode → Proceed with EnterPlanMode
-
-## ULTRAWORK Mode (Recommended)
-
-Include `ultrawork` or `ulw` keyword to activate maximum performance mode:
-
-```bash
-/vibe.run "feature-name" ultrawork   # All optimizations auto-enabled
-/vibe.run "feature-name" ulw         # Same (shorthand)
-```
-
-**Activated Features:**
-- Parallel sub-agent exploration (3+ concurrent)
-- **Background agents** - Prepare next Phase during implementation
-- **Phase pipelining** - Remove wait time between Phases
-- Boulder Loop (auto-continue until all Phases complete)
-- Auto-retry on error (max 3 times)
-- Auto-compress/save at 70%+ context
-- Continuous execution without confirmation between Phases
-
-**Speed Comparison:**
-
-| Mode | Per Phase | 5 Phases |
-|------|-----------|----------|
-| Sequential | ~2min | ~10min |
-| Parallel Exploration | ~1.5min | ~7.5min |
-| **ULTRAWORK Pipeline** | **~1min** | **~5min** |
-
-## Commands
-
-| Command | Description |
-|---------|-------------|
-| `/vibe.spec "feature-name"` | Write SPEC (PTCF structure) + parallel research |
-| `/vibe.run "feature-name"` | Execute implementation |
-| `/vibe.run "feature-name" ultrawork` | **Maximum performance mode** |
-| `/vibe.verify "feature-name"` | Verification |
-| `/vibe.review` | **Parallel code review** (13+ agents) |
-| `/vibe.reason "problem"` | Systematic reasoning |
-| `/vibe.analyze` | Project analysis |
-| `/vibe.utils --e2e` | E2E testing (Playwright) |
-| `/vibe.utils --diagram` | Generate diagrams |
-| `/vibe.utils --ui "description"` | UI preview |
-| `/vibe.utils --continue` | **Session restore** (load previous context) |
-
-## New Features (v2.1.0)
-
-### Parallel Code Review (/vibe.review)
-
-13+ specialized agents review code simultaneously:
-
-```
-┌─────────────────────────────────────────────────────────────────┐
-│  🚀 PARALLEL REVIEW AGENTS                                       │
-├─────────────────────────────────────────────────────────────────┤
-│  Security          │ security-reviewer, data-integrity-reviewer │
-│  Performance       │ performance-reviewer, complexity-reviewer  │
-│  Architecture      │ architecture-reviewer, simplicity-reviewer │
-│  Language-Specific │ python, typescript, rails, react reviewers │
-│  Context           │ git-history, test-coverage reviewers       │
-└─────────────────────────────────────────────────────────────────┘
-```
-
-**Priority System:**
-- 🔴 P1 (Critical): Blocks merge
-- 🟡 P2 (Important): Fix recommended
-- 🔵 P3 (Nice-to-have): Backlog
-
-### E2E Testing (/vibe.utils --e2e)
-
-Playwright-based automated testing:
-
-```bash
-/vibe.utils --e2e "login flow"   # Scenario test
-/vibe.utils --e2e --visual       # Visual regression test
-/vibe.utils --e2e --record       # Video recording
-```
-
-### Enhanced Research Agents
-
-4 parallel research agents run **after requirements confirmed** during `/vibe.spec`:
-
-```
-Q&A to confirm requirements → Parallel research → SPEC writing
-```
-
-| Agent | Role |
-|-------|------|
-| best-practices-agent | Best practices for confirmed feature+stack |
-| framework-docs-agent | Latest docs for confirmed stack (context7) |
-| codebase-patterns-agent | Analyze existing similar patterns |
-| security-advisory-agent | Security advisory for confirmed feature |
-
-**⚠️ Research runs after requirements confirmed** (VIBE principle: requirements first)
-
-## PTCF Structure
-
-SPEC documents are AI-executable prompt format:
-
-```
-<role>      AI role definition
-<context>   Background, tech stack, related code
-<task>      Phase-by-phase task list
-<constraints> Constraints
-<output_format> Files to create/modify
-<acceptance> Verification criteria
-```
-
-## Built-in Tools
-
-### Semantic Code Analysis
-| Tool | Purpose |
-|------|---------|
-| `vibe_find_symbol` | Find symbol definitions |
-| `vibe_find_references` | Find references |
-| `vibe_analyze_complexity` | Analyze complexity |
-| `vibe_validate_code_quality` | Validate quality |
-
-### Context Management
-| Tool | Purpose |
-|------|---------|
-| `vibe_start_session` | Start session (restore previous context) |
-| `vibe_auto_save_context` | Save current state |
-| `vibe_save_memory` | Save important decisions |
-
-## Agents
-
-### Review Agents (12)
-```
-.claude/agents/review/
-├── security-reviewer.md        # Security vulnerabilities
-├── performance-reviewer.md     # Performance bottlenecks
-├── architecture-reviewer.md    # Architecture violations
-├── complexity-reviewer.md      # Complexity exceeded
-├── simplicity-reviewer.md      # Over-abstraction
-├── data-integrity-reviewer.md  # Data integrity
-├── test-coverage-reviewer.md   # Missing tests
-├── git-history-reviewer.md     # Risk patterns
-├── python-reviewer.md          # Python specialist
-├── typescript-reviewer.md      # TypeScript specialist
-├── rails-reviewer.md           # Rails specialist
-└── react-reviewer.md           # React specialist
-```
-
-### Research Agents (4)
-```
-.claude/agents/research/
-├── best-practices-agent.md     # Best practices
-├── framework-docs-agent.md     # Framework docs
-├── codebase-patterns-agent.md  # Code pattern analysis
-└── security-advisory-agent.md  # Security advisory
-```
-
-## Skills
-
-### Git Worktree
-```bash
-# Isolated environment for PR review
-git worktree add ../review-123 origin/pr/123
-cd ../review-123 && npm test
-git worktree remove ../review-123
-```
-
-### Priority Todos
-```
-.claude/vibe/todos/
-├── P1-security-sql-injection.md   # 🔴 Blocks merge
-├── P2-perf-n1-query.md            # 🟡 Fix recommended
-└── P3-style-extract-helper.md     # 🔵 Backlog
-```
-
-## Context Management Strategy
-
-### Model Selection
-- **Exploration/Search**: Haiku (sub-agent default)
-- **Implementation/Debugging**: Sonnet
-- **Architecture/Complex logic**: Opus
-
-### At 70%+ Context (⚠️ Important)
-```
-❌ Don't use /compact (risk of information loss/distortion)
-✅ save_memory to store important decisions → /new for new session
-```
-
-vibe maintains context across sessions with its own memory system:
-1. `save_memory` - Explicitly save important decisions
-2. `/new` - Start new session
-3. `start_session` - Auto-restore previous session
-
-### Session Restore
-To continue previous work in a new session:
-```
-/vibe.utils --continue
-```
-This command calls `vibe_start_session` to restore previous context from project memory.
-
-### Other Commands
-- `/rewind` - Revert to previous point
-- `/context` - Check current usage
-
-### Using context7
-Use context7 plugin when you need latest library documentation:
-```
-"Search React 19 use() hook with context7"
-```
-
-## Documentation Guidelines
-
-### Diagrams/Structure Representation
-- Avoid ASCII boxes (┌─┐) → Alignment breaks with mixed-width characters
-- Use alternatives:
-  - Mermaid diagrams (GitHub/Notion supported)
-  - Markdown tables
-  - Indentation + separators
-
-### Preferred Formats
-| Purpose | Recommended |
-|---------|-------------|
-| Flowcharts | Mermaid flowchart |
-| Structure/Hierarchy | Indented lists |
-| Comparisons/Lists | Markdown tables |
-| Sequences | Mermaid sequenceDiagram |
-
-## Git Commit Rules
-
-**Must include:**
-- `.claude/vibe/specs/`, `.claude/vibe/features/`, `.claude/vibe/todos/` (project docs)
-- `.claude/vibe/config.json`, `.claude/vibe/constitution.md` (project config)
-- `CLAUDE.md`
-
-**Exclude (globally installed):**
-- `~/.claude/vibe/rules/`, `~/.claude/vibe/languages/`, `~/.claude/vibe/templates/` (global)
-- `~/.claude/commands/`, `~/.claude/agents/`, `~/.claude/skills/` (global)
-- `.claude/settings.local.json` (personal settings)
-
-## Getting Started
-
-```bash
-vibe init
-/vibe.spec "login feature"
-```
-
-## Full Workflow
-
-```mermaid
-flowchart TD
-    A["/vibe.spec"] --> B["(auto) SPEC review"]
-    B --> C["SPEC auto-enhancement"]
-    C --> D["/vibe.run ultrawork"]
-    D --> E["Gemini code review"]
-    E --> F["(auto) 13+ Agent Review"]
-    F --> G{"P1/P2 issues?"}
-    G -->|Yes| H["(auto) Auto-Fix"]
-    H --> I["✅ Done"]
-    G -->|No| I
-```
-
-| Step | Description | Automation |
-|------|-------------|------------|
-| 1. `/vibe.spec` | Collect requirements + Generate SPEC | Manual start |
-| 2. SPEC review | Gemini reviews SPEC + Auto-apply | ✅ Auto |
-| 3. `/vibe.run` | Implementation + Gemini review | Manual start |
-| 4. Agent Review | 13+ agent parallel review | ✅ Auto |
-| 5. Auto-Fix | P1/P2 issue auto-fix | ✅ Auto |
+# VIBE
+
+SPEC-driven AI Coding Framework (Claude Code Exclusive)
+
+## Response Language
+
+**IMPORTANT: Always respond in English unless the user explicitly requests otherwise.**
+
+## Response Language
+
+**IMPORTANT: Always respond in Korean (한국어) unless the user explicitly requests otherwise.**
+
+## Code Quality Standards (Mandatory)
+
+Follow these standards when writing code. See `~/.claude/vibe/rules/` (global) for detailed rules.
+
+### Core Principles
+- **Modify only requested scope** - Don't touch unrelated code
+- **Preserve existing style** - Follow project conventions
+- **Keep working code** - No unnecessary refactoring
+- **Respect user interrupts** - If user interrupts (Ctrl+C/Escape) and sends a new message, the previous task is CANCELLED. Do NOT resume or continue interrupted work. Respond ONLY to the new message.
+
+### Code Complexity Limits
+| Metric | Limit |
+|--------|-------|
+| Function length | ≤30 lines (recommended), ≤50 lines (allowed) |
+| Nesting depth | ≤3 levels |
+| Parameters | ≤5 |
+| Cyclomatic complexity | ≤10 |
+
+### TypeScript Rules
+- No `any` type → Use `unknown` + type guards
+- No `as any` casting → Define proper interfaces
+- No `@ts-ignore` → Fix type issues at root
+- Explicit return types on all functions
+
+### TypeScript Rules
+- No `any` type → use `unknown` + type guards
+- No `as any` casting → define proper interfaces
+- No `@ts-ignore` → fix type issues at root
+- Explicit return types on all functions
+
+### Error Handling Required
+- try-catch or error state required
+- Loading state handling
+- User-friendly error messages
+
+### Forbidden Patterns
+- No console.log in commits (remove after debugging)
+- No hardcoded strings/numbers → Extract to constants
+- No commented-out code in commits
+- No incomplete code without TODO
+
+## Workflow
+
+```
+/vibe.spec → (auto) SPEC review → /vibe.run → (auto) code review → (auto) fix → ✅ Done
+```
+
+**Automated Flow:**
+1. `/vibe.spec` - Write SPEC + **(auto)** Gemini review → Auto-apply
+2. `/vibe.run` - Implementation + Gemini review
+3. **(auto)** 13+ agent parallel review
+4. **(auto)** P1/P2 issue auto-fix
+
+## Plan Mode vs VIBE (Workflow Selection)
+
+**Offer choice to user on development requests:**
+
+| Task Size | Recommended |
+|-----------|-------------|
+| Simple changes (1-2 files) | Plan Mode |
+| Complex features (3+ files, research/verification needed) | `/vibe.spec` |
+
+| Item | Plan Mode | VIBE |
+|------|-----------|------|
+| Storage location | `~/.claude/plans/` (global) | `.claude/vibe/specs/` (project) |
+| Document format | Free form | PTCF structure (AI-optimized) |
+| Research | None | 4 parallel agents |
+| Verification | None | `/vibe.verify` against SPEC |
+| History | Not trackable | Git version control |
+
+**Rules:**
+- After `/vibe.analyze` or `/vibe.review` with dev/modify request → **Ask user for workflow choice**
+- User chooses VIBE → Wait for `/vibe.spec`
+- User chooses Plan Mode → Proceed with EnterPlanMode
+
+## ULTRAWORK Mode (Recommended)
+
+Include `ultrawork` or `ulw` keyword to activate maximum performance mode:
+
+```bash
+/vibe.run "feature-name" ultrawork   # All optimizations auto-enabled
+/vibe.run "feature-name" ulw         # Same (shorthand)
+```
+
+**Activated Features:**
+- Parallel sub-agent exploration (3+ concurrent)
+- **Background agents** - Prepare next Phase during implementation
+- **Phase pipelining** - Remove wait time between Phases
+- Boulder Loop (auto-continue until all Phases complete)
+- Auto-retry on error (max 3 times)
+- Auto-compress/save at 70%+ context
+- Continuous execution without confirmation between Phases
+
+**Speed Comparison:**
+
+| Mode | Per Phase | 5 Phases |
+|------|-----------|----------|
+| Sequential | ~2min | ~10min |
+| Parallel Exploration | ~1.5min | ~7.5min |
+| **ULTRAWORK Pipeline** | **~1min** | **~5min** |
+
+## Commands
+
+| Command | Description |
+|---------|-------------|
+| `/vibe.spec "feature-name"` | Write SPEC (PTCF structure) + parallel research |
+| `/vibe.run "feature-name"` | Execute implementation |
+| `/vibe.run "feature-name" ultrawork` | **Maximum performance mode** |
+| `/vibe.verify "feature-name"` | Verification |
+| `/vibe.review` | **Parallel code review** (13+ agents) |
+| `/vibe.reason "problem"` | Systematic reasoning |
+| `/vibe.analyze` | Project analysis |
+| `/vibe.utils --e2e` | E2E testing (Playwright) |
+| `/vibe.utils --diagram` | Generate diagrams |
+| `/vibe.utils --ui "description"` | UI preview |
+| `/vibe.utils --continue` | **Session restore** (load previous context) |
+
+## New Features (v2.5.7)
+
+### Intelligent Model Routing
+
+Automatic model selection based on task complexity:
+
+| Complexity | Model | When |
+|------------|-------|------|
+| Low (0-7) | Haiku | Simple fixes, searches |
+| Medium (8-19) | Sonnet | Standard features, 3-5 files |
+| High (20+) | Opus | Architecture, security, 6+ files |
+
+### Agent Tier System
+
+Cost-optimized agent variants:
+
+| Agent | Low | Medium | High |
+|-------|-----|--------|------|
+| explorer | explorer-low | explorer-medium | explorer |
+| implementer | implementer-low | implementer-medium | implementer |
+| architect | architect-low | architect-medium | architect |
+
+### Magic Keywords
+
+| Keyword | Effect |
+|---------|--------|
+| `ultrawork` / `ulw` | Parallel + auto-continue |
+| `ralph` | Persist until verified complete |
+| `ralplan` | Iterative planning + persistence |
+| `verify` | Strict verification mode |
+| `quick` | Fast mode, minimal verification |
+
+**Combinations supported:** `ralph ultrawork`, `ralph verify`, etc.
+
+### Skill Quality Gate
+
+Memory saves are validated for quality:
+
+- Rejects generic/searchable information
+- Requires context, specificity, actionability
+- Suggests principle format: "When X, do Y because Z"
+
+### HUD Status (Real-time)
+
+```bash
+node hooks/scripts/hud-status.js show full
+node hooks/scripts/hud-status.js start ultrawork "feature"
+node hooks/scripts/hud-status.js phase 2 5 "Implementing core"
+```
+
+### Pre/Post Tool Hooks
+
+- **PreToolUse**: Validates dangerous commands before execution
+- **PostToolUse**: Provides error recovery hints
+
+### Orchestrate Workflow
+
+Intent Gate → Codebase Assessment → Delegation → Verification pattern:
+
+```typescript
+import { checkIntentGate, assessCodebase, createDelegationPlan } from '@su-record/vibe/tools';
+```
+
+### UltraQA (5-Cycle Autonomous QA)
+
+```
+Test/Build/Lint → Fail → Architect Diagnosis → Executor Fix → Repeat (max 5)
+```
+
+Exit conditions: All pass, Max cycles, Same failure 3x
+
+### DeepInit (Hierarchical AGENTS.md)
+
+```
+project/
+├── AGENTS.md              ← Root
+├── src/
+│   └── AGENTS.md          ← <!-- Parent: ../AGENTS.md -->
+```
+
+### Skill Frontmatter System
+
+```yaml
+---
+name: my-skill
+model: sonnet
+triggers: [keyword1, keyword2]
+---
+```
+
+### Trigger-Based Skill Injection
+
+Skills in `~/.claude/vibe/skills/` or `.claude/vibe/skills/` auto-inject on keyword match.
+
+### Multi-Line HUD
+
+```bash
+node hooks/scripts/hud-multiline.js multi    # Tree view
+node hooks/scripts/hud-multiline.js compact  # 2-line view
+node hooks/scripts/hud-multiline.js single   # 1-line view
+```
+
+### Parallel Code Review (/vibe.review)
+
+13+ specialized agents review code simultaneously:
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  🚀 PARALLEL REVIEW AGENTS                                       │
+├─────────────────────────────────────────────────────────────────┤
+│  Security          │ security-reviewer, data-integrity-reviewer │
+│  Performance       │ performance-reviewer, complexity-reviewer  │
+│  Architecture      │ architecture-reviewer, simplicity-reviewer │
+│  Language-Specific │ python, typescript, rails, react reviewers │
+│  Context           │ git-history, test-coverage reviewers       │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+**Priority System:**
+- 🔴 P1 (Critical): Blocks merge
+- 🟡 P2 (Important): Fix recommended
+- 🔵 P3 (Nice-to-have): Backlog
+
+### E2E Testing (/vibe.utils --e2e)
+
+Playwright-based automated testing:
+
+```bash
+/vibe.utils --e2e "login flow"   # Scenario test
+/vibe.utils --e2e --visual       # Visual regression test
+/vibe.utils --e2e --record       # Video recording
+```
+
+### Enhanced Research Agents
+
+4 parallel research agents run **after requirements confirmed** during `/vibe.spec`:
+
+```
+Q&A to confirm requirements → Parallel research → SPEC writing
+```
+
+| Agent | Role |
+|-------|------|
+| best-practices-agent | Best practices for confirmed feature+stack |
+| framework-docs-agent | Latest docs for confirmed stack (context7) |
+| codebase-patterns-agent | Analyze existing similar patterns |
+| security-advisory-agent | Security advisory for confirmed feature |
+
+**⚠️ Research runs after requirements confirmed** (VIBE principle: requirements first)
+
+## PTCF Structure
+
+SPEC documents are AI-executable prompt format:
+
+```
+<role>      AI role definition
+<context>   Background, tech stack, related code
+<task>      Phase-by-phase task list
+<constraints> Constraints
+<output_format> Files to create/modify
+<acceptance> Verification criteria
+```
+
+## Built-in Tools
+
+### Semantic Code Analysis
+| Tool | Purpose |
+|------|---------|
+| `vibe_find_symbol` | Find symbol definitions |
+| `vibe_find_references` | Find references |
+| `vibe_analyze_complexity` | Analyze complexity |
+| `vibe_validate_code_quality` | Validate quality |
+
+### Context Management
+| Tool | Purpose |
+|------|---------|
+| `vibe_start_session` | Start session (restore previous context) |
+| `vibe_auto_save_context` | Save current state |
+| `vibe_save_memory` | Save important decisions |
+
+## Agents
+
+### Review Agents (12)
+```
+.claude/agents/review/
+├── security-reviewer.md        # Security vulnerabilities
+├── performance-reviewer.md     # Performance bottlenecks
+├── architecture-reviewer.md    # Architecture violations
+├── complexity-reviewer.md      # Complexity exceeded
+├── simplicity-reviewer.md      # Over-abstraction
+├── data-integrity-reviewer.md  # Data integrity
+├── test-coverage-reviewer.md   # Missing tests
+├── git-history-reviewer.md     # Risk patterns
+├── python-reviewer.md          # Python specialist
+├── typescript-reviewer.md      # TypeScript specialist
+├── rails-reviewer.md           # Rails specialist
+└── react-reviewer.md           # React specialist
+```
+
+### Research Agents (4)
+```
+.claude/agents/research/
+├── best-practices-agent.md     # Best practices
+├── framework-docs-agent.md     # Framework docs
+├── codebase-patterns-agent.md  # Code pattern analysis
+└── security-advisory-agent.md  # Security advisory
+```
+
+## Skills
+
+### Git Worktree
+```bash
+# Isolated environment for PR review
+git worktree add ../review-123 origin/pr/123
+cd ../review-123 && npm test
+git worktree remove ../review-123
+```
+
+### Priority Todos
+```
+.claude/vibe/todos/
+├── P1-security-sql-injection.md   # 🔴 Blocks merge
+├── P2-perf-n1-query.md            # 🟡 Fix recommended
+└── P3-style-extract-helper.md     # 🔵 Backlog
+```
+
+## Context Management Strategy
+
+### Model Selection
+- **Exploration/Search**: Haiku (sub-agent default)
+- **Implementation/Debugging**: Sonnet
+- **Architecture/Complex logic**: Opus
+
+### At 70%+ Context (⚠️ Important)
+```
+❌ Don't use /compact (risk of information loss/distortion)
+✅ save_memory to store important decisions → /new for new session
+```
+
+vibe maintains context across sessions with its own memory system:
+1. `save_memory` - Explicitly save important decisions
+2. `/new` - Start new session
+3. `start_session` - Auto-restore previous session
+
+### Session Restore
+To continue previous work in a new session:
+```
+/vibe.utils --continue
+```
+This command calls `vibe_start_session` to restore previous context from project memory.
+
+### Other Commands
+- `/rewind` - Revert to previous point
+- `/context` - Check current usage
+
+### Using context7
+Use context7 plugin when you need latest library documentation:
+```
+"Search React 19 use() hook with context7"
+```
+
+## Documentation Guidelines
+
+### Diagrams/Structure Representation
+- Avoid ASCII boxes (┌─┐) → Alignment breaks with mixed-width characters
+- Use alternatives:
+  - Mermaid diagrams (GitHub/Notion supported)
+  - Markdown tables
+  - Indentation + separators
+
+### Preferred Formats
+| Purpose | Recommended |
+|---------|-------------|
+| Flowcharts | Mermaid flowchart |
+| Structure/Hierarchy | Indented lists |
+| Comparisons/Lists | Markdown tables |
+| Sequences | Mermaid sequenceDiagram |
+
+## Git Commit Rules
+
+**Must include:**
+- `.claude/vibe/specs/`, `.claude/vibe/features/`, `.claude/vibe/todos/` (project docs)
+- `.claude/vibe/config.json`, `.claude/vibe/constitution.md` (project config)
+- `CLAUDE.md`
+
+**Exclude (globally installed):**
+- `~/.claude/vibe/rules/`, `~/.claude/vibe/languages/`, `~/.claude/vibe/templates/` (global)
+- `~/.claude/commands/`, `~/.claude/agents/`, `~/.claude/skills/` (global)
+- `.claude/settings.local.json` (personal settings)
+
+## Getting Started
+
+```bash
+vibe init
+/vibe.spec "login feature"
+```
+
+## Full Workflow
+
+```mermaid
+flowchart TD
+    A["/vibe.spec"] --> B["(auto) SPEC review"]
+    B --> C["SPEC auto-enhancement"]
+    C --> D["/vibe.run ultrawork"]
+    D --> E["Gemini code review"]
+    E --> F["(auto) 13+ Agent Review"]
+    F --> G{"P1/P2 issues?"}
+    G -->|Yes| H["(auto) Auto-Fix"]
+    H --> I["✅ Done"]
+    G -->|No| I
+```
+
+| Step | Description | Automation |
+|------|-------------|------------|
+| 1. `/vibe.spec` | Collect requirements + Generate SPEC | Manual start |
+| 2. SPEC review | Gemini reviews SPEC + Auto-apply | ✅ Auto |
+| 3. `/vibe.run` | Implementation + Gemini review | Manual start |
+| 4. Agent Review | 13+ agent parallel review | ✅ Auto |
+| 5. Auto-Fix | P1/P2 issue auto-fix | ✅ Auto |