Gestion de cycle PDCA avec agents intégrés

VérifiéPrudence

Gère le cycle PDCA d'un projet via des agents intégrés (CTO, PO, Tech Lead, etc.) avec retours bidirectionnels. Utilise les chemins de documents de bkit.config.json. Déclenche les phases research, plan, design, do, analyze, iterate/report. Utile pour automatiser et structurer les boucles de rétroaction entre les rôles projet.

Spar Skills Guide Bot
DeveloppementAvancé
02/06/2026
Claude Code
#pdca#agent-workflow#project-management#bidirectional-feedback#development-cycle

Recommandé pour

Synthèse automatique

Orchestre le cycle PDCA (Plan-Do-Check-Act) pour un projet logiciel en utilisant un réseau d'agents IA spécialisés avec des boucles de rétroaction bidirectionnelles.

Points forts

  • Automatisation de workflows multi-agents complexes
  • Phases structurées avec génération de documents
  • Boucles de validation intégrées (CTO, PO, Tech Lead, etc.)
  • Séparation claire entre intention (what/why) et implémentation (how) dans les documents

Limites

  • Nécessite une configuration projet détaillée via bkit.config.json
  • Peut être excessif pour des changements simples ou rapides
  • Dépend de mappings d'agents spécifiques qui doivent être maintenus
Quand l'utiliser

Lors de la gestion de fonctionnalités complexes ou d'améliorations nécessitant une collaboration multi-agents et un suivi structuré.

Quand l'éviter

Pour des modifications mineures ou non critiques où un agent unique suffit.

Analyse de sécurité

Prudence
Score qualité90/100

The skill is a PDCA workflow manager that legitimately requires Bash, Task, and file write operations. It does not instruct destructive or exfiltrating actions, but the presence of powerful execution tools warrants caution.

Points d'attention
  • •Uses Bash tool, which could execute arbitrary shell commands if the skill is invoked with untrusted input or if the instructions are manipulated.
  • •Uses Task tool to invoke downstream agents, which may have their own capabilities; the chain of tool usage could escalate privileges.
  • •Skill writes files and modifies `.pdca-status.json`, which could affect project tracking if misused.
Comment cette analyse est faite

Exemples

Start PDCA cycle
pdca Implement user authentication
Plan a feature
plan Add payment processing
Get current status
status

name: pdca description: | Project-level PDCA skill override for gaegulzip. Integrates with project agents (CTO, PO, Tech Lead, etc.) and uses platform-specific document paths from bkit.config.json.

All agent interactions are BIDIRECTIONAL — downstream agents can push back to upstream agents when issues are found, with structured feedback loops.

Agent workflow per phase (↔ = bidirectional):

  • Research: clarify ↔ research-director → CTO 통합 평가 (아키텍처+복잡도, BLOCK → clarify 재호출)
  • Plan: PO ↔ CTO (타당성 스캔, BLOCK → PO 재호출) → 사용자 승인 → CTO 라우팅 ↔ PO (Scope Mismatch → PO 보완 + 재승인)
  • Design: ui-ux-designer ↔ tech-lead (Design Pushback 루프) + API Contract 양방향 검증 (server/frontend tech-lead)
  • Do: CTO (work-plan) → developers ↔ CTO (BLOCKED:QUESTIONS → 라우팅 → 답변 → 재개)
  • Analyze: gap-detector ↔ CTO (integration review)
  • Iterate/Report: bkit agents

Triggers: pdca, research, plan, design, analyze, report, status, next, iterate argument-hint: "[action] [feature]" user-invocable: true agents: research-brainstorm: clarify (Skill) research-director: bkit/research-director research-feasibility-cto: cto (Feasibility Council)

plan: product-owner plan-review: interactive-review:review (Skill) feasibility: cto routing: cto design-server: server/tech-lead design-mobile-ui: mobile/ui-ux-designer design-mobile-visual: frontend-design (Skill) design-mobile-tech: mobile/tech-lead design-web-ui: web/ui-ux-designer design-web-visual: frontend-design (Skill) design-web-tech: web/tech-lead do-distribute: cto do-server: server/node-developer do-mobile: mobile/flutter-developer do-web: web/react-developer test-scenario: test-scenario-generator design-review: plan-review (Skill, MANDATORY) analyze: bkit:gap-detector code-review: plan-review (Skill, MANDATORY) review: cto iterate: bkit:pdca-iterator report: bkit:report-generator verify: independent-reviewer allowed-tools:

  • Read
  • Write
  • Edit
  • Glob
  • Grep
  • Bash
  • Task
  • TaskCreate
  • TaskUpdate
  • TaskList
  • AskUserQuestion imports:
  • ${PLUGIN_ROOT}/templates/plan.template.md
  • ${PLUGIN_ROOT}/templates/design.template.md
  • ${PLUGIN_ROOT}/templates/do.template.md
  • ${PLUGIN_ROOT}/templates/analysis.template.md
  • ${PLUGIN_ROOT}/templates/report.template.md
  • ${PLUGIN_ROOT}/templates/iteration-report.template.md next-skill: null pdca-phase: null task-template: "[PDCA] {feature}"

PDCA Skill (gaegulzip Project Override)

Unified Skill for managing PDCA cycle with agent-integrated workflow. Each PDCA phase automatically invokes the appropriate project agents from bkit.config.json. Phase 상세 절차는 phases/ 디렉토리의 개별 파일 참조.

Config Reference

Read bkit.config.json to determine:

  • context.defaultLanguage → 모든 출력(문서, 상태 메시지, 에이전트 응답)에 이 언어 사용
  • pdca.planDocPaths → document output paths
  • pdca.designDocPaths → document output paths
  • pdca.analyzeDocPaths → analysis document output paths
  • pdca.reportDocPaths → report document output paths
  • platforms.{platform}.agents → agent mapping per phase
  • pdca.statusFile → .pdca-status.json path
  • pdca.matchRateThreshold → 90 (default)
  • pdca.maxIterations → 5 (default)

Language Rule

All output MUST be in the language specified by bkit.config.json → context.defaultLanguage. When calling agents via Task tool, always append to the prompt: \nIMPORTANT: Respond and write all documents in {defaultLanguage}. Code, paths, and technical terms stay in English.

⚠️ Intent-Only Document Rule (Plan & Design 문서 필수 원칙)

Plan/Design 문서는 "무엇을(WHAT)"과 "왜(WHY)"만 작성한다. "어떻게(HOW)" 즉 구현은 절대 작성하지 않는다.

이 원칙을 위반하는 예시:

  • ❌ 파일명/경로 지정: src/services/authService.ts에 구현
  • ❌ 코드 구조 지시: useEffect로 상태를 관리하고 Redux store에 저장
  • ❌ 함수 시그니처: async function validateToken(token: string): Promise<boolean>
  • ❌ DB 쿼리 작성: SELECT * FROM users WHERE email = ?
  • ❌ 특정 라이브러리 사용법: zod.object({ email: z.string().email() })로 검증

이 원칙을 지키는 예시:

  • ✅ 사용자 행동 정의: 사용자는 이메일과 비밀번호로 로그인할 수 있다
  • ✅ 비즈니스 규칙: 비밀번호는 최소 8자, 영문+숫자+특수문자 포함
  • ✅ 데이터 요구사항: 사용자 프로필에는 닉네임, 프로필 이미지, 가입일이 포함된다
  • ✅ API 계약 (Design): POST /auth/login → 요청: { email, password }, 응답: { token, user }
  • ✅ 제약조건: 토큰 만료 시간은 24시간

이유: 구현은 Do 단계의 개발자 에이전트가 결정한다. 설계 문서에 구현을 쓰면 에이전트 자율성을 제한하고, 코드 변경 시 문서 부패를 유발한다.

모든 Plan/Design phase 에이전트 프롬프트에 이 원칙을 포함시킬 것.

Product Naming Convention

{product}는 앱 이름을 그대로 사용합니다:

  • wowa — 메인 앱
  • admin — 어드민 대시보드
  • talmosang — 탈모상 AI 두피 분석

경로 예: docs/wowa/auth/user-story.md, docs/admin/dashboard/web-brief.md

Fullstack frontendType 제약

frontendType은 단일 값만 허용합니다 ("mobile" 또는 "web"). Mobile + Web 동시 개발이 필요한 경우, 별도 PDCA 사이클로 분리합니다:

  • 예: wowa feature = fullstack (frontendType: mobile), admin feature = fullstack (frontendType: web)

Status Tracking

.pdca-status.json stores platform and phase per feature:

{
  "features": {
    "{feature}": {
      "phase": "research | plan | design | do | analyze | iterate | completed | archived",
      "platform": "server | mobile | web | fullstack",
      "frontendType": "mobile | web",
      "startedAt": "ISO timestamp",
      "documents": {}
    }
  }
}

IMPORTANT:

  • platform is determined during Plan phase by CTO and reused in all subsequent phases.
  • frontendType is set when platform is fullstack to distinguish Mobile vs Web frontend agents.

Arguments

| Argument | Description | Example | |----------|-------------|---------| | research [feature] | clarify → research-director | /pdca research user-auth | | plan [feature] | PO → interactive-review → CTO routing | /pdca plan user-auth | | design [feature] | Design docs (per platform) | /pdca design user-auth | | do [feature] | CTO distribution → dev agents | /pdca do user-auth | | analyze [feature] | Gap analysis + CTO review | /pdca analyze user-auth | | iterate [feature] | Auto improvement | /pdca iterate user-auth | | report [feature] | Completion report | /pdca report user-auth | | archive [feature] | Archive documents | /pdca archive user-auth | | cleanup [feature] | Cleanup archived | /pdca cleanup | | status | Current status | /pdca status | | next | Next phase guide | /pdca next |


Phase Details

각 phase 실행 시 해당 파일을 Read하여 상세 절차를 따릅니다.

| Phase | File | Summary | |-------|------|---------| | research | phases/research.md | clarify → research-director (Decision Points) → CTO 통합 평가 (아키텍처+복잡도, BLOCK↔clarify 재호출) | | plan | phases/plan.md | clarify → PO (Research Gate) ↔ CTO 타당성 스캔 (BLOCK→PO 재호출) → 사용자 승인 → CTO 라우팅 ↔ PO (Scope Mismatch→보완+재승인) | | design | phases/design.md | 플랫폼별 디스패치 + tech-lead ↔ designer Pushback 루프 + API Contract 양방향 검증 (Fullstack) | | do | phases/do.md | 선행조건 검증 → 설계 리뷰 → CTO work-plan → 사용자 승인 → 실행 모드 → 에이전트 구현 ↔ BLOCKED:QUESTIONS 프로토콜 | | analyze | phases/analyze.md | gap-detector → FINDINGS 구조화 → 코드 리뷰 → CTO 통합 리뷰 | | iterate | phases/iterate.md | FINDINGS severity 순 에이전트 디스패치 또는 Ralph Loop 자율 수정 | | report | phases/report.md | report-generator → 배포/유지보수/v2 가이드 → CHANGELOG | | archive | phases/archive.md | 문서 아카이브 → status 업데이트 |


status (Status Check)

  1. Read .pdca-status.json
  2. Display: feature, phase, platform, matchRate, documents
  3. Visualize progress
📊 PDCA Status
─────────────────────────────
Feature: {feature}
Platform: {platform}
Phase: {phase}
Match Rate: {matchRate}%
─────────────────────────────
[Research] ✅ → [Plan] ✅ → [Design] ✅ → [Do] 🔄 → [Check] ⏳ → [Act] ⏳

next (Next Phase)

| Current | Next | Action | |---------|------|--------| | None | research (권장) 또는 plan | research는 선택적. 복잡한 기능은 research 권장, 단순 기능은 plan 직행 가능 | | research | plan | PO → interactive-review → CTO routing | | plan | design | Platform-based design agents | | design | do | CTO distribution → dev agents | | do | check | Gap analysis + CTO review | | check (<90%) | act | Auto-iterate | | check (>=90%) | report | Completion report | | report | archive | Archive documents |


Agent Integration Summary (All Bidirectional)

↔ = 양방향 (피드백 경로 있음), ──→ = 단방향 (최종 출력) 참조: .claude/guide/agent-communication-protocol.md — 전 phase 양방향 통신 프로토콜

Research: clarify (user) ──→ research-director (research.md + Decision Points)
                                       │
                                       ↓ CTO 통합 평가 (아키텍처 + 복잡도)
                                       │
                              BLOCK ↔ clarify 재호출 (max 2회)
                              CONCERN → AskUserQuestion
                                       │
                                       ↓ Decision Points → PO에게 전달
Plan:    PO (user-story.md) ↔ CTO (⓪-pre 타당성 스캔)
              │                        │ BLOCK → PO 재호출 (max 2회)
              │                        ↓ PASS/WARN
         interactive-review (사용자 승인) ──→ CTO (⓪ 플랫폼 라우팅)
                                                   │
                                       Scope Mismatch ↔ PO 보완 + 사용자 재승인
                                                   ↓ platform & frontendType stored
Design:  ┌── Server: tech-lead (server-brief)
         ├── Mobile: ui-ux-designer ↔ tech-lead (Design Pushback 루프, max 2회)
         │           └── frontend-design (Skill) 포함
         └── Web:    ui-ux-designer ↔ tech-lead (Design Pushback 루프, max 2회)
                     └── frontend-design (Skill) 포함
              │      Fullstack = Server + frontend (by frontendType)
              │      + API Contract 양방향 검증 (server tech-lead ↔ CTO ↔ frontend tech-lead)
              │
Do:      CTO (work-plan) ──→ developers (node/flutter/react)
              │                    ↔ BLOCKED:QUESTIONS → CTO 라우팅 → 답변자 → 재개
              │              OR: Ralph Loop (단일 에이전트 자율 반복)
              │
Analyze: gap-detector ↔ CTO (review)
              │
Iterate: pdca-iterator (if < 90%)  OR: Ralph Loop (FINDINGS 자율 수정)
              │
Report:  report-generator ──→ CHANGELOG
              │
Verify:  independent-reviewer (optional)

Bidirectional Communication Reference

전 phase의 양방향 통신 프로토콜 상세는 아래 문서를 참조합니다:

  • 통신 프로토콜: .claude/guide/agent-communication-protocol.md
  • phase별 상세: phases/{phase}.md

CTO Involvement (5 Points)

| Phase | CTO Role | Output | 양방향 | |-------|---------|--------|--------| | Research (after research-director) | 아키텍처+복잡도 통합 평가 | PASS/CONCERN/BLOCK + complexity 판정 | BLOCK → clarify 재호출 | | Plan (after PO, before user review) | ⓪-pre 타당성 스캔 | PASS/WARN/BLOCK 판정 | BLOCK → PO 재호출 | | Plan (after user review) | ⓪ 플랫폼 라우팅 + Scope Mismatch 감지 | platform in status | Mismatch → PO 보완 + 재승인 | | Do (before devs) | ② 작업 분배 + BLOCKED:QUESTIONS 라우팅 | work-plan.md (플랫폼별) | BLOCKED → 적절한 에이전트 라우팅 | | Analyze (after gap) | ③ 통합 리뷰 | cto-review.md (플랫폼별) | — |

Du fichier au premier essai

Le téléchargement contient exactement le contenu affiché sur cette fiche. Cette copie peut différer de la dernière version du dépôt source.

Consulter la source et ses fichiers annexes

Ce téléchargement contient uniquement SKILL.md. Les scripts, références, modèles et dépendances éventuels ne sont pas inclus. Récupérez les fichiers nécessaires depuis la source en conservant leur arborescence avant de poursuivre.

1. Télécharger le fichier

Télécharger SKILL.md

Depuis la racine d’un projet de test, enregistrez le fichier sous .skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md, ou utilisez la commande suivante. Ce dossier sert à examiner la skill avant son installation.

Commande pour macOS/Linux ou WSL : un shell POSIX et curl sont nécessaires. Sous Windows sans WSL, utilisez le bouton de téléchargement et créez le dossier manuellement.

mkdir -p '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i' && curl --fail --show-error --location 'https://skills-guides.com/api/skills/pdca-cycle-management-with-integrated-agents-mm5t8a1i/download' --output '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md.part' && mv '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md.part' '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md'

2. Vérifier les prérequis

Lisez SKILL.md avant de l’utiliser. Si le fichier cite des scripts ou services externes, examinez-les et préparez les accès nécessaires dans votre environnement de test.

Aucune section de prérequis reconnue dans ce fichier. Cela ne signifie pas qu’il est autonome : les bibliothèques, outils et accès peuvent être cités ailleurs dans le contenu ou dans le dépôt source.

Les plateformes de la fiche sont des indications du catalogue. Les chemins ci-dessous suivent la documentation des outils ; ils ne prouvent pas que cette skill y a été exécutée. Aucune version minimale de ces outils n’est attestée pour cette fiche.

3. Charger la skill

Après vérification, copiez le dossier complet dans le projet avec l’une des commandes ci-dessous, puis ouvrez une nouvelle session de l’outil. La commande s’arrête si un dossier du même nom existe déjà : comparez les fichiers avant de le remplacer.

Claude Code
test -f '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md' && test -s '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md' && mkdir -p '.claude/skills' && test ! -e '.claude/skills/pdca' && cp -R '.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i' '.claude/skills/pdca'

OpenSkills

OpenSkills doit déjà être installé. La procédure locale install/read est contrôlée avec OpenSkills 1.5.0, qui exige Node.js ≥ 20.6.0. Aucun slug du catalogue n’est résolu automatiquement.

Installation officielle d’OpenSkills
openskills install './.skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i'

openskills read 'pdca-cycle-management-with-integrated-agents-mm5t8a1i'

La lecture doit afficher le contenu du fichier installé et son dossier de base. Elle ne lance ni la skill ni ses dépendances.

Pour un agent qui lit AGENTS.md, synchronisez ensuite l’index. Cette commande modifie AGENTS.md : relisez le diff avant de l’utiliser.

openskills sync

Mode manuel ou autre outil : joignez SKILL.md à une conversation, ou ouvrez-le dans un agent capable de lire les fichiers. Ajoutez les ressources annexes nécessaires. Il n’existe pas de dossier universel reconnu par tous les outils.

4. Faire un premier essai

Utilisez une entrée courte et fictive : un endpoint d’API, un petit document ou un brief d’interface. Dans une nouvelle conversation, joignez le fichier ou donnez son chemin à l’agent, puis adaptez cette demande :

Lis .skill-downloads/pdca-cycle-management-with-integrated-agents-mm5t8a1i/SKILL.md et applique ses instructions à cet exemple : [décrire une entrée de test]. Indique les fichiers, dépendances ou accès qui manquent. Produis le résultat demandé par la skill sans inventer de données ni exécuter de script non examiné.

Résultat attendu : une sortie correspondant aux instructions de la skill, ou une liste précise des prérequis manquants. Une simple confirmation de lecture ne valide pas le premier usage. Si une ressource externe est indisponible, ce parcours reste à tester.

Skills similaires