Back to Skills

moai-alfred-rules

modu-ai
Updated Today
32 views
424
78
424
View on GitHub
Metaaiautomation

About

This skill defines essential rules and standards for Alfred SuperAgent development based on November 2025 enterprise specifications. It provides governance for workflow validation, quality gates, and architecture compliance including 3-layer architecture, 4-step workflow, and TRUST 5 quality standards. Developers should use it to ensure their agent implementations follow established patterns for skill invocation, TAG chain integrity, and MoAI-ADK compliance.

Quick Install

Claude Code

Recommended
Plugin CommandRecommended
/plugin add https://github.com/modu-ai/moai-adk
Git CloneAlternative
git clone https://github.com/modu-ai/moai-adk.git ~/.claude/skills/moai-alfred-rules

Copy and paste this command in Claude Code to install this skill

Documentation

Skill 개요

moai-alfred-rules는 Alfred SuperAgent의 의사결정과 실행을 제어하는 핵심 프레임워크입니다.

항목
버전4.0.0 (November 2025 enterprise)
티어Alfred (상위 계층)
자동 로드규칙 검증, 품질 게이트, 아키텍처 규칙 필요 시
아키텍처 패러다임Agent-First (Command → Agent → Skill → Hook)
워크플로우 모델4-Step ADAP Workflow

무엇을 하는가?

핵심 책임

  1. 3-Layer Architecture 정의: Commands → Agents → Skills 계층 분리
  2. 4-Step Workflow 규칙: ADAP (Analyze, Design, Assure, Produce) + Intent
  3. Agent-First Paradigm: 모든 실행 작업을 agents에 위임
  4. Skill 호출 규칙: 10+ mandatory patterns, invocation syntax
  5. AskUserQuestion 패턴: 5가지 필수 사용 시나리오
  6. TRUST 5 Quality Gates: T/R/U/S/T 각각의 검증 기준
  7. TAG Chain Integrity: SPEC→TEST→CODE→DOC 추적
  8. Commit Message Standards: TDD cycle message formats

언제 사용하는가?

필수 시나리오 (MUST use)

상황사용 여부
✅ Skill() 호출 규칙 검증필수
✅ Command vs Agent 역할 분리필수
✅ AskUserQuestion 사용 판단필수
✅ TRUST 5 준수 확인필수
✅ TAG 체인 무결성 검증필수
✅ 커밋 메시지 형식 확인필수
✅ 워크플로우 compliance 검증필수
✅ Agent delegation 올바른지 확인필수
✅ 품질 게이트(quality gate) 통과필수
✅ 아키텍처 규칙 위반 감지필수

Rule 1: 3-Layer Architecture (November 2025 Standard)

계층 구조

┌─────────────────────────────────────┐
│ Commands (Orchestration Only)       │ ← User-facing entry points
│ /alfred:0-project, /alfred:1-plan   │   No direct execution
│ /alfred:2-run, /alfred:3-sync       │
└──────────┬──────────────────────────┘
           │ Task(subagent_type="...")
           ↓
┌─────────────────────────────────────┐
│ Agents (Domain Expertise)           │ ← Deep reasoning
│ spec-builder, tdd-implementer       │   Complex decisions
│ test-engineer, doc-syncer           │   Plan → Execute
│ git-manager, qa-validator           │
└──────────┬──────────────────────────┘
           │ Skill("skill-name")
           ↓
┌─────────────────────────────────────┐
│ Skills (Knowledge Capsules)         │ ← Reusable patterns
│ 55 specialized Skills               │   Playbooks
│ < 1000 lines each                   │   Best practices
└─────────────────────────────────────┘

규칙 1.1: Commands - Orchestration ONLY

금지 사항 (❌):

# ❌ WRONG: 직접 작업 실행
echo "Building application..."
python setup.py build
git commit -m "Build"

# ❌ WRONG: Skill 직접 호출
Skill("moai-alfred-rules")  # Commands에서 금지!

# ❌ WRONG: 복잡한 로직 구현
if feature_type == "backend":
  # 복잡한 비즈니스 로직...

필수 사항 (✅):

# ✅ CORRECT: Agent에 위임
Task(
  subagent_type="tdd-implementer",
  description="Build and test application",
  prompt="Implement feature with RED-GREEN-REFACTOR cycle"
)

# ✅ CORRECT: 의사결정만 하고 위임
if user_approval_needed:
  AskUserQuestion(...)  # 사용자 확인 후 위임
  Task(subagent_type="implementation-planner", ...)

규칙 1.2: Agents - Domain Expertise Ownership

agent의 책임:

  • ✅ 복잡한 분석 & 추론 (deep reasoning)
  • ✅ 계획 수립 (planning)
  • ✅ 의사결정 (decision-making)
  • ✅ Skill 호출 및 조율 (orchestration within domain)
  • ✅ 작업 실행 (execution)

예시: tdd-implementer Agent:

Agent receives: "Implement user authentication"
  ↓
1. Analyze: SPEC 검토, 요구사항 분석
2. Design: 아키텍처 설계
3. RED: test-engineer Skill 호출 → 테스트 작성
4. GREEN: 코드 구현
5. REFACTOR: 최적화
6. Commit: git-manager 호출 → commit
  ↓
Returns: Fully tested, documented, committed code

규칙 1.3: Skills - Knowledge Capsules (Stateless)

Skill 특성:

  • ✅ 상태가 없음 (stateless)
  • ✅ 재사용 가능 (reusable)
  • ✅ Agent에 의해 호출됨 (called by agents)
  • ✅ 1000줄 이하 (< 1000 lines)
  • ✅ 단일 주제 (single topic)

금지 (❌):

# ❌ WRONG: Skill이 다른 Skill 호출
Skill("moai-foundation-git")  # ← Skills에서 금지!

# ❌ WRONG: Skill이 Task() 실행
Task(subagent_type="...")  # ← Skills에서 금지!

# ❌ WRONG: Skill이 상태 유지
state = {"counter": 0}  # ← Stateless 위반

Rule 2: 4-Step Agent-Based Workflow (November 2025)

Phase Overview

Phase 0: INTENT                      Phase 1: ANALYZE
┌──────────────────┐               ┌──────────────────┐
│ User Request     │ ─clarity?─→   │ WebSearch        │
│ Ambiguous?       │ NO             │ WebFetch         │
│ ✅ YES: clarify  │     YES        │ Research         │
│ ✅ NO: continue  │────────────→   │ Best practices   │
└──────────────────┘               └──────────────────┘
                                          ↓
Phase 3: ASSURE                     Phase 2: DESIGN
┌──────────────────┐               ┌──────────────────┐
│ Quality Gate     │←──────────────→│ Architecture     │
│ TRUST 5          │                │ Latest info      │
│ TAG integrity    │                │ Version specs    │
│ Compliance       │                │ Design patterns  │
└──────────────────┘               └──────────────────┘
        ↓
Phase 4: PRODUCE
┌──────────────────┐
│ Skill invocation │
│ File generation  │
│ Commit           │
└──────────────────┘

Phase 0: Intent (사용자 의도 파악)

규칙 0.1: Intent가 모호하면 AskUserQuestion 사용

상황: "데이터 처리 모듈 만들어줘"

Step 1: Intent 평가
  ├─ Clarity: LOW (어떤 데이터? 어떤 처리?)
  └─ Action: AskUserQuestion 사용

Step 2: 명확화
AskUserQuestion({
  question: "어떤 데이터를 처리하나요?",
  options: [
    "CSV 파일",
    "데이터베이스",
    "API 응답"
  ]
})

Step 3: 명확한 요구사항 확보 후 다음 phase로

Phase 1: Analyze (정보 수집 & 연구)

도구: WebSearch, WebFetch

Task 1: Research latest information
  ├─ Search: "[framework] [version] best practices 2025"
  ├─ Fetch: Official documentation URLs
  └─ Validate: Cross-check multiple sources

Task 2: Collect best practices
  ├─ Official docs
  ├─ Industry standards
  └─ Current patterns (2025)

Task 3: Identify version-specific guidance
  ├─ Current stable version
  ├─ Breaking changes
  └─ Deprecation warnings

Phase 2: Design (아키텍처 설계)

입력: Phase 1의 연구 결과 출력: November 2025 최신 정보를 기반한 설계

Design Activities:
  ├─ 최신 정보 기반 이름 지정
  ├─ 현재 버전 명시
  ├─ 최신 패턴 포함
  ├─ 공식 문서 링크
  └─ 마지막 업데이트 날짜 포함

Phase 3: Assure (품질 검증)

TRUST 5 Quality Gates:

Gate검증 기준
Test85%+ coverage, 모든 경로 테스트
ReadableClean code, SOLID 원칙, 주석 포함
Unified일관된 패턴, 중복 제거, 네이밍 표준
SecuredOWASP Top 10 확인, 비밀정보 제거

Phase 4: Produce (생성 & 배포)

책임: Skill 호출 (예: moai-skill-factory)

Actions:
  ├─ 템플릿 적용
  ├─ 파일 생성
  ├─ 메타데이터 추가
  ├─ 최신 정보 embedded
  ├─ 공식 문서 링크 포함
  └─ Version date 표기

Rule 3: Agent-First Paradigm (Critical Enforcement)

금지된 작업 (❌ ABSOLUTELY FORBIDDEN)

Alfred (또는 Command)가 직접 실행하면 절대 안 됩니다:

  1. 직접 bash 명령 실행

    # ❌ WRONG
    bash("git commit -m 'message'")
    os.system("python build.py")
    
    # ✅ CORRECT
    Task(subagent_type="git-manager", prompt="Commit changes")
    
  2. 파일 읽기/쓰기

    # ❌ WRONG
    with open("file.py", "w") as f:
        f.write(code)
    
    # ✅ CORRECT
    Task(subagent_type="file-manager", prompt="Create file")
    
  3. Git 직접 조작

    # ❌ WRONG
    subprocess.run(["git", "push", "origin", "main"])
    
    # ✅ CORRECT
    Task(subagent_type="git-manager", prompt="Push changes")
    
  4. 코드 분석 직접 실행

    # ❌ WRONG
    lines = len(open("file.py").readlines())
    
    # ✅ CORRECT
    Task(subagent_type="code-analyzer", prompt="Analyze code")
    
  5. 테스트 직접 실행

    # ❌ WRONG
    subprocess.run(["pytest", "tests/"])
    
    # ✅ CORRECT
    Task(subagent_type="test-engineer", prompt="Run tests")
    

의무 위임 (✅ MANDATORY DELEGATION)

작업위임 대상패턴
계획 수립plan-agentTask(subagent_type="plan-agent", ...)
코드 개발tdd-implementerTask(subagent_type="tdd-implementer", ...)
테스트 작성test-engineerTask(subagent_type="test-engineer", ...)
문서화doc-syncerTask(subagent_type="doc-syncer", ...)
Git 작업git-managerTask(subagent_type="git-manager", ...)
품질 검증qa-validatorTask(subagent_type="qa-validator", ...)
사용자 질문ask-user-questionsAskUserQuestion(...)

Rule 4: 10 Mandatory Skill Invocations

규칙 4.1: Skill Invocation Pattern

Syntax:

Skill("skill-name")  # Explicit invocation only

10가지 필수 Skill

#Skill용도Invocation
1moai-foundation-trustTRUST 5 검증Skill("moai-foundation-trust")
2moai-foundation-tagsTAG 검증 & 추적Skill("moai-foundation-tags")
3moai-foundation-specsSPEC 작성 & 검증Skill("moai-foundation-specs")
4moai-foundation-earsEARS 요구사항 형식Skill("moai-foundation-ears")
5moai-foundation-gitGit 워크플로우Skill("moai-foundation-git")
6moai-foundation-langs언어 & 스택 감지Skill("moai-foundation-langs")
7moai-essentials-debug디버깅 & 에러 분석Skill("moai-essentials-debug")
8moai-essentials-refactor리팩토링 & 개선Skill("moai-essentials-refactor")
9moai-essentials-perf성능 최적화Skill("moai-essentials-perf")
10moai-essentials-review코드 리뷰 & 품질Skill("moai-essentials-review")

규칙 4.2: Skill Invocation 예제

# Context 1: TRUST 5 검증 필요
if validation_required:
    Skill("moai-foundation-trust")
    # → 반환: TRUST score, violations, recommendations

# Context 2: TAG 체인 검증
if tag_integrity_check:
    Skill("moai-foundation-tags")
    # → 반환: 고아 TAG, 깨진 체인, 제안

# Context 3: SPEC 작성
if spec_needed:
    Skill("moai-foundation-specs")
    # → 반환: SPEC template, validation results

# Context 4: Git 워크플로우
if git_decision_needed:
    Skill("moai-foundation-git")
    # → 반환: branch strategy, commit format, merge rules

Rule 5: AskUserQuestion Patterns (5가지 필수 시나리오)

규칙 5.1: MANDATORY Scenarios

Scenario 1: 기술 스택 모호

상황: "Python 웹 프레임워크 추천해줄래?"

AskUserQuestion({
  question: "어떤 유형의 애플리케이션?",
  header: "Application Type",
  options: [
    { label: "REST API", description: "High performance APIs" },
    { label: "Web Application", description: "Traditional MVC" },
    { label: "Microservice", description: "Event-driven" }
  ]
})

Scenario 2: 아키텍처 결정

상황: "데이터베이스 모델을 어떻게 설계?"

AskUserQuestion({
  question: "어떤 데이터 특성?",
  header: "Data Model",
  options: [
    { label: "Relational", description: "Structured, ACID" },
    { label: "Document", description: "Flexible schema" },
    { label: "Graph", description: "Relationships" }
  ]
})

Scenario 3: 의도 모호

상황: "코드 개선해줄래?"

AskUserQuestion({
  question: "어떤 측면을 개선?",
  header: "Improvement Focus",
  options: [
    { label: "Performance", description: "Speed & efficiency" },
    { label: "Readability", description: "Code clarity" },
    { label: "Security", description: "Vulnerability fixes" }
  ],
  multiSelect: true  # 복수 선택 허용
})

Scenario 4: 기존 컴포넌트 영향

상황: "패키지 업그레이드하려는데 호환성?"

AskUserQuestion({
  question: "기존 코드 호환성 유지 필요?",
  header: "Compatibility",
  options: [
    { label: "Full compatibility", description: "Maintain all APIs" },
    { label: "Deprecation path", description: "Gradual migration" },
    { label: "Breaking OK", description: "Version bump allowed" }
  ]
})

Scenario 5: 자원 제약

상황: "시스템 리팩토링 계획 수립"

AskUserQuestion({
  question: "예상 개발 기간?",
  header: "Timeline",
  options: [
    { label: "1 week", description: "Focused scope" },
    { label: "2-4 weeks", description: "Medium refactor" },
    { label: "1+ months", description: "Comprehensive" }
  ]
})

규칙 5.2: 올바른 사용법

❌ WRONG (평문 질문):

사용자: "뭘 선호해?"
응답: "음... 생각해봤는데 아마도..."

✅ CORRECT (AskUserQuestion):

AskUserQuestion({
  question: "어떤 접근을 선호하시나요?",
  header: "Approach",
  multiSelect: false,
  options: [
    { label: "Option A", description: "Benefit A, Cost B" },
    { label: "Option B", description: "Benefit C, Cost D" }
  ]
})

Rule 6: TRUST 5 Quality Gates (November 2025 Enterprise)

각 Gate의 검증 기준

T: Test First (85%+ Coverage)

requirements:
  coverage: "≥ 85%"
  coverage_tools: ["pytest-cov", "coverage.py"]
  test_types:
    - unit_tests: "각 함수/메서드"
    - integration_tests: "모듈 간 상호작용"
    - edge_cases: "경계값, 에러 처리"
  
validation:
  - All code paths executed
  - Error conditions tested
  - Mock external dependencies
  - No skipped tests (×skip, ×xfail)

R: Readable (Clean Code)

requirements:
  code_standards:
    - SOLID 원칙 준수
    - DRY (Don't Repeat Yourself)
    - KISS (Keep It Simple, Stupid)
  
  documentation:
    - Function docstrings
    - Complex logic comments
    - Type hints (Python 3.10+)
  
  naming:
    - Descriptive variable names
    - Consistent conventions
    - No single-letter vars (except i, j, k in loops)

U: Unified (Consistent Patterns)

requirements:
  consistency:
    - Same patterns across codebase
    - No duplicate logic
    - Shared utilities for common tasks
  
  conventions:
    - Consistent indentation (4 spaces)
    - Consistent naming (snake_case, PascalCase)
    - Consistent import order
  
  validation:
    - Linting (flake8, black, isort)
    - Static analysis (pylint, mypy)
    - No code duplication (DRY violations)

S: Secured (OWASP Top 10)

requirements:
  security_checks:
    - No hardcoded credentials
    - No SQL injection vectors
    - No XXE vulnerabilities
    - Input validation
    - Output encoding
  
  tools:
    - bandit (Python security linter)
    - safety (dependency vulnerabilities)
    - SAST scanning
  
  validation:
    - No secrets committed
    - Dependencies scanned
    - Known CVEs checked
requirements:
  tag_chain:
    - Complete history traceability
  
  documentation:
  
  validation:
    - Bidirectional references

Rule 7: TAG Chain Integrity Rules

규칙 7.1: TAG Naming Convention

Format: @<DOMAIN>-<###>

Examples:

규칙 7.2: TAG Lifecycle


  └─ Example: test_auth.py - test_auth_success()

  └─ Example: auth.py - authenticate_user()

  └─ Example: CHANGELOG.md

규칙 7.3: TAG Validation Rules

❌ 위반:

# 1. Orphan TAG (TEST/CODE 없음)

# 2. 깨진 체인 (missing step)

# 3. 불일치 (다른 번호)

✅ 정확:

# Complete chain

Rule 8: Commit Message Standards (TDD Cycle)

규칙 8.1: Commit Format

Format:
<type>(<tag>): <subject>

<body>

<footer>

규칙 8.2: TDD Cycle Commits

RED Commit (테스트 작성):


- test_successful_login()
- test_invalid_credentials()
- test_expired_token()

Status: RED (Tests fail as expected)

GREEN Commit (구현):


- Implement authenticate_user() function
- Add token generation
- Add error handling

Status: GREEN (All tests pass)

REFACTOR Commit (최적화):


- Extract token validation to separate function
- Add caching for user lookups
- Improve error messages

Status: PASSING (Tests still pass, code improved)

규칙 8.3: Commit 타입 (Conventional Commits)

타입설명예시
chore빌드/설정chore: Update dependencies

Rule 9: Workflow Compliance Validation

규칙 9.1: Compliance Checklist

Before Commit:

  • 테스트 통과 (85%+ coverage)
  • Linting 통과 (black, flake8)
  • Security scan 통과 (bandit, safety)
  • 비밀정보 제거 (no secrets)
  • 커밋 메시지 형식 맞음

Before Merge:

  • TAG 체인 완전
  • TRUST 5 통과
  • Code review 완료
  • CI/CD 통과
  • PR 설명 포함

규칙 9.2: Violation Response

Violation Detected:

1. 자동 탐지 (hook)
2. 사용자에게 알림
3. 수정 요청
4. 재검증
5. Pass/Fail 결정

3-Level Progressive Disclosure

Level 1: 빠른 시작 (Beginner - 10분)

당신이 알아야 할 것:

  1. Command = 조율만, Agent = 실행
  2. Skill = 전문 도구
  3. 규칙 위반 → 에러

Level 2: 실무 패턴 (Intermediate - 30분)

당신이 해야 할 것:

  1. AskUserQuestion 사용 시점 판단
  2. TRUST 5 검증
  3. TAG 체인 유지
  4. 커밋 메시지 형식

Level 3: 심화 (Advanced - 1시간)

당신이 최적화할 것:

  1. Agent delegation 전략
  2. Skill 조합 활용
  3. Workflow 자동화
  4. 규칙 예외 관리

공식 자료 & 참고 링크

Architecture References

  • Command-Agent-Skill 패턴: Internal moai-adk documentation
  • ADAP Workflow: Internal workflow definition

Quality Standards

  • TRUST 5 Framework: Skill("moai-foundation-trust")
  • TAG System: Skill("moai-foundation-tags")
  • Commit Standards: Skill("moai-foundation-git")

Enterprise Standards (November 2025)

  • Agent-First Architecture: InfoQ (agentic-ai-architecture-framework)
  • TDD Best Practices: Official pytest documentation
  • Code Quality: OWASP Top 10, SOLID principles

버전: 4.0.0 (November 2025 Enterprise Standard) 최종 업데이트: 2025-11-12 유지보수자: GoosLab (Alfred SuperAgent Framework)

GitHub Repository

modu-ai/moai-adk
Path: src/moai_adk/templates/.claude/skills/moai-alfred-rules
agentic-aiagentic-codingagentic-workflowclaudeclaudecodevibe-coding

Related Skills

sglang

Meta

SGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.

View skill

evaluating-llms-harness

Testing

This Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.

View skill

content-collections

Meta

This skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.

View skill

llamaguard

Other

LlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.

View skill