@musashishao/agent-kit 1.9.0 → 1.9.1

package/.agent/skills/vibe-coding-guard/SKILL.md

@@ -0,0 +1,328 @@
+---
+name: vibe-coding-guard
+description: "Enforce Vibe Coding 2026 best practices. Combines VIBES Framework, CONDUCTOR pattern, and trust spectrum for comprehensive AI development governance."
+version: "1.0.0"
+---
+
+# 🎸 Vibe Coding Guard
+
+> **The complete governance framework for AI-assisted development**
+
+This skill enforces all Vibe Coding 2026 best practices, integrating VIBES Framework, CONDUCTOR responsibilities, and the Trust Spectrum into a cohesive workflow.
+
+---
+
+## When to Use This Skill
+
+- Any AI-assisted development session
+- Large codegen tasks
+- Multi-agent workflows
+- Production-critical code
+- Teaching vibe coding practices
+
+---
+
+## 1. The Vibe Coding Spectrum
+
+```
+PURE VIBE                                              RESPONSIBLE VIBE
+(Throwaway)                                            (Production)
+    │                                                       │
+    ▼                                                       ▼
+┌───────────────────────────────────────────────────────────────────┐
+│                                                                   │
+│  Blind Trust     Light Review     Deep Review     Full TDD       │
+│      │               │                │               │          │
+│  Weekend         Prototypes       Standard        Critical       │
+│  Projects        & MVPs           Features        Systems        │
+│                                                                   │
+│  Risk: High     Risk: Medium     Risk: Low       Risk: Min      │
+└───────────────────────────────────────────────────────────────────┘
+```
+
+---
+
+## 2. VIBES Pre-Flight Checklist
+
+Before ANY AI code generation:
+
+```markdown
+## VIBES Pre-Flight
+
+### V - Verification Plan
+- [ ] What command will prove success?
+- [ ] What tests need to pass?
+- [ ] What evidence will you require?
+
+### I - Intent Captured
+- [ ] Explicit request documented
+- [ ] Implicit needs identified
+- [ ] Unknown needs anticipated
+- [ ] Anti-goals stated
+
+### B - Bifurcation Points
+- [ ] Technical decisions identified
+- [ ] Architecture choices listed
+- [ ] Scope boundaries defined
+- [ ] Trade-offs documented
+
+### E - Emergence Preparedness
+- [ ] Scope limits defined
+- [ ] Dependency policy set
+- [ ] Pattern expectations established
+
+### S - Safety Layers Active
+- [ ] Input validation ready
+- [ ] Security scanner configured
+- [ ] Review process defined
+- [ ] Rollback capability confirmed
+```
+
+---
+
+## 3. CONDUCTOR Mode
+
+Developer role in vibe coding mode:
+
+```
+YOU ARE THE CONDUCTOR 🎼
+━━━━━━━━━━━━━━━━━━━━━━
+
+Your job is NOT to write code.
+Your job is to:
+
+1. CONTEXT    - Feed rich context to agents
+2. ORCHESTRATE - Direct the workflow
+3. NAVIGATE   - Guide through decisions
+4. DECOMPOSE  - Break down complexity
+5. UNDERSTAND - Know the domain deeply
+6. COORDINATE - Sync multiple agents
+7. TRUST      - Build verified trust
+8. OBSERVE    - Watch for emergence
+9. REFINE     - Iterate to perfection
+```
+
+---
+
+## 4. Session Governance
+
+### Start of Session
+
+```python
+def start_vibe_session(task: Task) -> Session:
+    session = Session()
+    
+    # 1. Classify task risk
+    session.risk_level = classify_risk(task)
+    
+    # 2. Set verification level
+    session.verification = VERIFICATION_MAP[session.risk_level]
+    
+    # 3. Load relevant trust scores
+    session.agent_trust = load_agent_trust()
+    
+    # 4. Initialize emergence detector
+    session.emergence = EmergenceDetector()
+    
+    # 5. Set up decision logger
+    session.decisions = DecisionMemory()
+    
+    return session
+```
+
+### During Session
+
+```python
+def process_ai_output(output: Output, session: Session) -> Result:
+    # 1. Emergence check
+    emergence = session.emergence.check(output)
+    if emergence.has_issues:
+        return Result.NEEDS_REVIEW(emergence.issues)
+    
+    # 2. Hallucination check
+    hallucinations = check_hallucinations(output)
+    if hallucinations:
+        return Result.REJECT(hallucinations)
+    
+    # 3. Trust-based verification
+    verification_level = session.get_verification_level()
+    if not verify(output, verification_level):
+        return Result.REJECT("Verification failed")
+    
+    # 4. Update trust
+    session.agent_trust.update(outcome="success")
+    
+    return Result.ACCEPT(output)
+```
+
+### End of Session
+
+```python
+def end_vibe_session(session: Session):
+    # 1. Log all decisions
+    session.decisions.persist()
+    
+    # 2. Update trust scores
+    persist_trust_scores(session.agent_trust)
+    
+    # 3. Generate session summary
+    summary = generate_summary(session)
+    
+    # 4. Store for future reference
+    store_session(session, summary)
+```
+
+---
+
+## 5. Risk Classification
+
+| Project Type | Risk Level | Verification | Trust Required |
+|--------------|------------|--------------|----------------|
+| Weekend hack | Low | Light | Zero Trust OK |
+| Prototype | Low-Medium | Standard | Calculated |
+| Internal tool | Medium | Standard | Calculated |
+| Customer-facing | High | Deep | Earned |
+| Payment/Banking | Critical | Full TDD | Earned + Review |
+| Healthcare/Legal | Critical | Full + Audit | Max + External |
+
+---
+
+## 6. Quick Reference Cards
+
+### The 5-Second VIBES Check
+
+```
+V - What's my verification command?
+I - Did I capture the full intent?
+B - Any decisions I need to make?
+E - Am I watching for unexpected output?
+S - What safety layers are active?
+```
+
+### The CONDUCTOR Quick Check
+
+```
+✓ Did I provide enough context?
+✓ Am I directing, not writing?
+✓ Did I guide through choices?
+✓ Is complexity broken down?
+✓ Do I understand the domain?
+✓ Are agents coordinated?
+✓ Is trust verified, not assumed?
+✓ Am I watching for emergence?
+✓ Will I iterate to refine?
+```
+
+---
+
+## 7. Integration with Workflows
+
+### /create Workflow
+
+```
+1. VIBES Pre-Flight Check
+2. Intent Capture (Socratic)
+3. Bifurcation Analysis
+4. Agent Orchestration
+5. Emergence Detection
+6. Verification Gate
+7. Decision Logging
+```
+
+### /debug Workflow
+
+```
+1. Context Gathering (CONDUCTOR-C)
+2. Root Cause Analysis
+3. Fix Proposal (with Bifurcation)
+4. Implementation
+5. Emergence Check
+6. Verification (fresh evidence)
+```
+
+### /orchestrate Workflow
+
+```
+1. VIBES Pre-Flight
+2. Multi-Agent Trust Check
+3. Task Decomposition (CONDUCTOR-D)
+4. Parallel Agent Execution
+5. Emergence Detection per Agent
+6. Cross-Agent Verification
+7. Synthesis with Safety Layers
+```
+
+---
+
+## 8. Enforcement Levels
+
+| Level | Description | Enforcement |
+|-------|-------------|-------------|
+| **Advisory** | Suggest best practices | Recommendations only |
+| **Standard** | Require pre-flight | Block without VIBES check |
+| **Strict** | Full compliance | Block on any violation |
+| **Paranoid** | Maximum safety | Require human approval |
+
+```python
+def set_enforcement_level(project_type: str) -> EnforcementLevel:
+    return {
+        "prototype": EnforcementLevel.ADVISORY,
+        "internal": EnforcementLevel.STANDARD,
+        "production": EnforcementLevel.STRICT,
+        "critical": EnforcementLevel.PARANOID
+    }.get(project_type, EnforcementLevel.STANDARD)
+```
+
+---
+
+## 9. Metrics & Reporting
+
+### Session Metrics
+
+```json
+{
+  "session_id": "2026-01-29-001",
+  "duration_minutes": 45,
+  "vibes_pre_flights": 3,
+  "emergence_detections": 1,
+  "hallucinations_caught": 0,
+  "decisions_logged": 5,
+  "verification_passes": 8,
+  "verification_failures": 1,
+  "trust_updates": [
+    {"agent": "coder", "delta": +2},
+    {"agent": "reviewer", "delta": +1}
+  ]
+}
+```
+
+### Weekly Report
+
+```markdown
+## Vibe Coding Report: Week 4
+
+### Summary
+- Sessions: 15
+- Emergence Issues: 3 (caught before commit)
+- Hallucinations: 1 (phantom package)
+- Trust Score Changes: +12 total
+
+### Top Decisions
+1. Chose PostgreSQL over MongoDB (ADR-004)
+2. Implemented feature flags (ADR-005)
+
+### Learnings
+- Agent X struggles with auth code (lower trust)
+- VIBES pre-flight caught 3 scope creep issues
+```
+
+---
+
+## 10. Related Skills
+
+- `verification-gate` - V in VIBES
+- `brainstorming` - I in VIBES (Intent Capture)
+- `emergence-detector` - E in VIBES
+- `trust-spectrum` - Trust management
+- `decision-memory` - Decision logging
+- `anti-hallucination` - Hallucination prevention

package/.agent/templates/AGENTS.game.md

@@ -0,0 +1,63 @@
+# AGENTS.md - Game Development Team
+
+> **Role:** Game Studio Orchestration Map
+> **Status:** Active
+
+---
+
+## 🎮 Core Game Team
+
+| Agent | Role | Focus |
+|-------|------|-------|
+| `game-developer` | **Lead Developer** | Gameplay logic, engine integration (Unity/Godot), mechanics |
+| `unity-mobile-master` | **Mobile Specialist** | Optimization, touch controls, mobile build pipeline |
+| `ai-asset-factory` | **Art Director** | 2D/3D asset generation, textures, UI elements |
+| `ai-audio-factory` | **Audio Engineer** | BGM, SFX, voice generation, adaptive audio |
+| `project-planner` | **Producer** | Task tracking, milestones, GDD management |
+
+---
+
+## 🛠️ Workflows (Slash Commands)
+
+| Command | Purpose |
+|---------|---------|
+| `/gamekit-spec` | Create Game Design Document (GDD) & Specs |
+| `/gamekit-plan` | Generate technical implementation plan |
+| `/gamekit-tasks` | Create executable task list |
+| `/gamekit-qa` | Playtesting & Quality Assurance loop |
+| `/gamekit-launch` | Build & Store submission workflow |
+
+---
+
+## 📂 Directory Structure
+
+```
+.
+├── Assets/                 # Game Assets (Unity/Godot standard)
+├── ProjectSettings/        # Engine settings
+├── Docs/                   # Documentation
+│   ├── GDD.md              # Game Design Document
+│   ├── SPEC-*.md           # Feature Specs
+│   └── PLAN-*.md           # Implementation Plans
+└── .agent/                 # Agent Configuration
+    ├── memory/             # Session & Decision logs
+    └── skills/             # Project-specific skills
+```
+
+---
+
+## 📝 Conventions
+
+- **Game Loop:** Logic in `Update()` (variable) vs `FixedUpdate()` (physics).
+- **Assets:** Use `Resources/` sparingly. Prefer Addressables/AssetBundles.
+- **Code Style:** C# (Unity) or GDScript (Godot) follows standard guidelines.
+- **Commit:** Git output must exclude huge binaries (use LFS).
+
+---
+
+## 🚀 Quick Start
+
+1. **Initialize:** Run `/gamekit-init` to setup structure.
+2. **Spec:** Run `/gamekit-spec` to define your game idea.
+3. **Plan:** Run `/gamekit-plan` to create technical roadmap.
+4. **Build:** Use `@game-developer` to start coding.

package/.agent/templates/docs/WORKFLOW_GUIDE.en.md

@@ -0,0 +1,100 @@
+# Agent Kit Workflow Guide
+
+This guide provides details on how to use workflows in Agent Kit according to the **Vibe Coding 2026** standard.
+
+## 🌐 Localization
+
+All workflow descriptions below will automatically change according to your `lang` setting.
+- To change language to Vietnamese: `ak memory set-lang --value vi`
+- After that, when you type `/` in chat, suggestions will be displayed in Vietnamese.
+
+---
+
+## 🛠️ Workflow List & Use Cases
+
+### 1. Planning Group
+Used to shape the project before implementation.
+
+| Command | Use Case | Main Action |
+|:---|:---|:---|
+| `/spec` | For complex new features | Create `SPEC-*.md` (Requirements Specification) |
+| `/plan` | After Spec or for large tasks | Create `PLAN-*.md` (Implementation Plan) |
+| `/brainstorm` | When needing ideas or solutions | Provide multiple options with pros/cons |
+| `/context` | When context is too large/cluttered | Optimize file list for AI focus |
+
+### 2. Execution Group
+Used to write and update code.
+
+| Command | Use Case | Main Action |
+|:---|:---|:---|
+| `/create` | Start new project/module | Write code following TDD (Red-Green-Refactor) |
+| `/enhance` | Update, add small features | Safely modify existing code |
+| `/orchestrate`| Extremely difficult, multi-layered tasks | Mobilize >3 specialized Agents in parallel |
+| `/ui-ux-pro-max`| Focused on UI/UX | Apply 50+ modern styles (Glassmorphism, etc.) |
+
+### 3. Quality & Debug Group
+Used to maintain code quality.
+
+| Command | Use Case | Main Action |
+|:---|:---|:---|
+| `/debug` | For bugs or strange behavior | Analyze Root Cause and propose fix/prevention |
+| `/test` | When testing is needed | Create unit/integration tests or run coverage |
+| `/autofix` | Terminal command failure | Analyze error and attempt fix (max 3 tries) |
+| `/quality` | When AI becomes "verbose" | Optimize context and instructions for accuracy |
+
+### 4. Ops & Status Group
+Used to manage progress and deployment.
+
+| Command | Use Case | Main Action |
+|:---|:---|:---|
+| `/status` | Check progress | Show task status dashboard and Agent trust |
+| `/dashboard` | Visual interface | Open project dashboard (if supported) |
+| `/deploy` | Push code to production | Run Pre-deploy checklist and execute deploy |
+| `/next` | When "stuck" | Suggest next step based on Plan and AGENTS.md |
+
+### 5. Specialty Group
+Used for specific types of projects.
+
+| Command | Use Case | Main Action |
+|:---|:---|:---|
+| `/gamekit-*` | Game Development | 6 specialized workflows for Game Development |
+| `/saas` | Build SaaS product | Landing page, Stripe, Subscription |
+| `/marketing` | Growth planning | Copywriting, Funnel, Campaign |
+| `/pentest` | Security check | Scan vulnerabilities, check secret leaks |
+| `/ai-agent` | Build AI Agent | Design automated Agent systems |
+
+---
+
+## 🔄 Persistence: Memory Storage
+
+**Question:** *Will Memory be overwritten during init or Kit update?*
+
+**Answer:** **NO.**
+
+Your data is absolutely protected by:
+1. **Existence Check**: The constructor in `memory_manager.py` always checks `if not path.exists()`. It only creates default files if you don't have them.
+2. **Safe Update**: When upgrading Agent Kit, the system only updates "skills" and "workflows". Data in `.agent/memory/` belongs to the project.
+3. **Goal**: Ensure your "Decision Memory" is always inherited across sessions.
+
+---
+
+## 💻 CLI Commands
+
+| Command | Description |
+|:---|:---|
+| `ak init` | Initialize AI Infrastructure for the project |
+| `ak status` | View AI Infra status |
+| `ak sync` | Sync graph.json and RAG chunks |
+| `ak workflows` | Show available workflows guide |
+| `ak memory get --type brain` | View project "brain" |
+| `ak memory log-decision` | Record rationale for a technical decision |
+| `ak memory set-lang --value en` | Change system language |
+
+---
+
+## 🚀 Quick Start
+
+1. **Install**: `npm install @musashishao/agent-kit`
+2. **Initialize**: `ak init`
+3. **View guide**: `ak workflows`
+4. **Start**: Type `/spec` in AI chat to define your first feature!

package/.agent/templates/docs/WORKFLOW_GUIDE.vi.md

@@ -0,0 +1,100 @@
+# Agent Kit Workflow Guide
+
+Hướng dẫn này chi tiết về cách sử dụng các workflow trong Agent Kit theo tiêu chuẩn **Vibe Coding 2026**.
+
+## 🌐 Đa ngôn ngữ (Localization)
+
+Tất cả các mô tả workflow bên dưới sẽ tự động thay đổi theo cài đặt `lang` của bạn.
+- Để đổi ngôn ngữ sang tiếng Việt: `ak memory set-lang --value vi`
+- Khi đó, khi bạn gõ `/` trong chat, các gợi ý sẽ hiển thị bằng tiếng Việt.
+
+---
+
+## 🛠️ Danh sách Workflow & Trường hợp sử dụng
+
+### 1. Nhóm Lập kế hoạch (Planning)
+Dùng để định hình dự án trước khi thực hiện.
+
+| Lệnh | Trường hợp sử dụng | Hành động chính |
+|:---|:---|:---|
+| `/spec` | Khi có tính năng mới phức tạp | Tạo file `SPEC-*.md` (Đặc tả yêu cầu) |
+| `/plan` | Sau khi có Spec hoặc cho task lớn | Tạo file `PLAN-*.md` (Kế hoạch thực hiện) |
+| `/brainstorm` | Khi cần ý tưởng hoặc giải pháp | Đưa ra nhiều option kèm ưu/nhược điểm |
+| `/context` | Khi context quá lớn/rối | Tối ưu hóa file list để AI tập trung hơn |
+
+### 2. Nhóm Thực thi (Execution)
+Dùng để viết và cập nhật code.
+
+| Lệnh | Trường hợp sử dụng | Hành động chính |
+|:---|:---|:---|
+| `/create` | Bắt đầu dự án/module mới | Viết code theo chuẩn TDD (Red-Green-Refactor) |
+| `/enhance` | Cập nhật, thêm tính năng nhỏ | Chỉnh sửa code hiện có một cách an toàn |
+| `/orchestrate`| Task cực khó, đa tầng | Huy động >3 Agent chuyên biệt làm việc song song |
+| `/ui-ux-pro-max`| Chuyên về giao diện (UI/UX) | Áp dụng 50+ styles hiện đại (Glassmorphism, v.v.) |
+
+### 3. Nhóm Kiểm soát lỗi (Quality & Debug)
+Dùng để duy trì chất lượng code.
+
+| Lệnh | Trường hợp sử dụng | Hành động chính |
+|:---|:---|:---|
+| `/debug` | Khi có lỗi hoặc hành vi lạ | Phân tích Root Cause và đề xuất cách sửa/phòng ngừa |
+| `/test` | Khi cần kiểm thử | Tạo unit test, integration test hoặc chạy coverage |
+| `/autofix` | Lệnh terminal bị lỗi | Tự động phân tích lỗi và thử sửa (max 3 lần) |
+| `/quality` | Khi AI bắt đầu "lan man" | Tối ưu lại context và chỉ dẫn để tăng độ chính xác |
+
+### 4. Nhóm Vận hành (Ops & Status)
+Dùng để quản lý tiến độ và triển khai.
+
+| Lệnh | Trường hợp sử dụng | Hành động chính |
+|:---|:---|:---|
+| `/status` | Muốn biết mình đang ở đâu | Hiện Dashboard trạng thái task và độ tin cậy Agent |
+| `/dashboard` | Giao diện trực quan | Mở Dashboard dự án (nếu hỗ trợ) |
+| `/deploy` | Đưa code lên production | Chạy Pre-deploy checklist và thực thi lệnh deploy |
+| `/next` | Khi bị "kẹt" | Gợi ý bước đi tiếp theo dựa trên Plan và AGENTS.md |
+
+### 5. Nhóm Chuyên biệt (Specialty)
+Dùng cho các loại dự án đặc thù.
+
+| Lệnh | Trường hợp sử dụng | Hành động chính |
+|:---|:---|:---|
+| `/gamekit-*` | Phát triển Game | 6 workflows chuyên dụng cho Game Development |
+| `/saas` | Xây dựng sản phẩm SaaS | Landing page, Stripe, Subscription |
+| `/marketing` | Lập kế hoạch tăng trưởng | Copywriting, Funnel, Campaign |
+| `/pentest` | Kiểm tra bảo mật | Quét lỗ hổng, kiểm tra rò rỉ secret |
+| `/ai-agent` | Xây dựng AI Agent | Thiết kế hệ thống Agent tự động |
+
+---
+
+## 🔄 Persistence: Lưu trữ Memory
+
+**Câu hỏi:** *Memory có bị ghi đè khi init hoặc update Kit không?*
+
+**Trả lời:** **KHÔNG.**
+
+Dữ liệu của bạn được bảo vệ tuyệt đối nhờ các cơ chế sau:
+1. **Kiểm tra tồn tại**: Hàm khởi tạo trong `memory_manager.py` luôn kiểm tra `if not path.exists()`. Nó chỉ tạo file mặc định nếu bạn chưa có.
+2. **Cập nhật an toàn**: Khi bạn nâng cấp Agent Kit, hệ thống chỉ cập nhật phần "kỹ năng" (skills) và "quy trình" (workflows). Dữ liệu trong thư mục `.agent/memory/` là của riêng dự án.
+3. **Mục tiêu**: Đảm bảo "Decision Memory" (Lịch sử quyết định) của bạn luôn được kế thừa qua các phiên làm việc.
+
+---
+
+## 💻 Lệnh CLI Hỗ trợ
+
+| Lệnh | Mô tả |
+|:---|:---|
+| `ak init` | Khởi tạo AI Infrastructure cho dự án |
+| `ak status` | Xem trạng thái AI Infra |
+| `ak sync` | Đồng bộ graph.json và RAG chunks |
+| `ak workflows` | Hiển thị danh sách workflows có sẵn |
+| `ak memory get --type brain` | Xem "bộ não" của dự án |
+| `ak memory log-decision` | Ghi lại lý do cho một quyết định kỹ thuật |
+| `ak memory set-lang --value vi` | Đổi ngôn ngữ toàn hệ thống |
+
+---
+
+## 🚀 Quick Start
+
+1. **Cài đặt**: `npm install @musashishao/agent-kit`
+2. **Khởi tạo**: `ak init`
+3. **Xem hướng dẫn**: `ak workflows`
+4. **Bắt đầu**: Gõ `/spec` trong AI chat để định nghĩa tính năng đầu tiên!

package/.agent/workflows/ai-agent.md

@@ -1,5 +1,7 @@
 ---
 description: Build AI agent workflow. From system design to autonomous execution loops.
+description_en: Build AI agent workflow. From system design to autonomous execution loops.
+description_vi: Xây dựng workflow cho AI Agent. Từ thiết kế hệ thống đến các vòng lặp thực thi tự trị.
 ---
 
 # 🤖 /ai-agent - AI Agent Builder Workflow

package/.agent/workflows/autofix.md

@@ -1,6 +1,7 @@
 ---
 description: Autonomous self-healing loop for failed commands.
 description_vi: Vòng lặp tự phục hồi tự động cho các lệnh bị lỗi.
+description_en: Autonomous self-healing loop for failed commands.
 ---
 
 # /autofix - Autonomous Self-Healing Loop

package/.agent/workflows/brainstorm.md

@@ -1,6 +1,7 @@
 ---
 description: Structured brainstorming for projects and features. Explore multiple options before implementation.
 description_vi: Brainstorming có cấu trúc cho dự án và tính năng. Khám phá nhiều phương án trước khi triển khai.
+description_en: Structured brainstorming for projects and features. Explore multiple options before implementation.
 ---
 
 # /brainstorm - Structured Idea Exploration

package/.agent/workflows/context.md

@@ -1,6 +1,7 @@
 ---
 description: Automatically generate optimal context for complex tasks using Context Engineering strategies.
 description_vi: Tự động tạo ngữ cảnh tối ưu cho các tác vụ phức tạp bằng chiến lược Context Engineering.
+description_en: Automatically generate optimal context for complex tasks using Context Engineering strategies.
 ---
 
 # Context Optimization Workflow

package/.agent/workflows/create.md

@@ -1,6 +1,7 @@
 ---
 description: Command to create a new application. Activates the App Builder skill and initiates user dialogue.
 description_vi: Tạo ứng dụng mới với kiến trúc AI-Ready. Kích hoạt kỹ năng App Builder và bắt đầu đối thoại với người dùng.
+description_en: Command to create a new application. Activates the App Builder skill and initiates user dialogue.
 ---
 
 # /create - Create Application
@@ -15,15 +16,29 @@ This command starts a new application creation process with **AI-Ready Infrastru
 
 ### Steps:
 
-1. **Phase 0: Quality Gate** ⭐
+## 💾 Git Checkpoint (Safety First)
+
+Before starting work:
+1. [ ] **Clean working tree?** (Run `git status`)
+2. [ ] **On correct branch?** (Not `main` or `master`)
+3. [ ] **Latest pulled?** (Run `git pull`)
+
+> 🔴 **STOP** if git state is dirty. Ask user to commit or stash first.
+
+---
+
+### Steps:
+
+1. **Phase 0: VIBES Pre-Flight** ⭐ NEW:
+   - **Verification**: Run `ak memory set --key last_verification --value "[CMD]"` with the command that proves success.
+   - **Intent**: Use `@[skills/intent-capture]` to extract explicit/implicit needs and anti-goals.
+   - **Bifurcation**: Use `@[skills/bifurcation-analysis]` to map technical vs architectural branches.
+   - **Emergence**: Set scope boundaries to prevent creep.
+   - **Safety**: Verify code rules compliance.
+
+2. **Phase 1: Quality Gate** ⭐
    - **SDD Check**: Ensure `SPEC-*.md` exists and is marked as `Approved` for complex features (3+ files).
    - **SDD Check**: Ensure `PLAN-*.md` exists and aligns with the Spec.
-   - **TDD Red-Phase Check** ⭐ NEW:
-     1. Generate at least 1 failing test based on the approved Spec (e.g., `tests/[feature].test.ts`).
-     2. Run the test: `npm test` or equivalent.
-     3. Verify the test **FAILS** (exit code != 0).
-     4. **If test passes or doesn't exist**: **HALT** and notify user. The test must fail first (RED phase) to ensure the Spec is correctly understood.
-   - **Action**: If missing, REDIRECT to `/spec`, `/plan`, or `/test red [feature]` before proceeding.
 
 2. **Request Analysis**
    - Understand what the user wants
@@ -124,7 +139,23 @@ If request is unclear, ask these questions:
 - What are the basic features?
 - Who will use it?
 
-Use defaults, add details later.
+> 🔴 **NEVER assume. Wait for user answers before proceeding.**
+
+---
+
+## 📝 Decision Logging (MANDATORY)
+
+After EVERY clarifying question is answered, log to `.agent/memory/decisions.md`:
+
+```markdown
+## Session: [YYYY-MM-DD HH:MM]
+
+### Q: [Question asked]
+**A:** [User's answer]
+**Impact:** [How this affects implementation]
+```
+
+> ⚠️ If `decisions.md` doesn't exist, create it first.
 
 ---