AI agent skill · by @ww-w-ai
Pdca
Unified skill for managing the entire PDCA cycle. Supports Plan → Design → Do → Check → Act workflow with automatic phase progression. Use proactively when user mentions PDCA cycle, planning, design documents, gap analysis, iteration, or completion reports. Triggers: pdca, plan, design, analyze, check, report, status, next, iterate, 계획, 설계, 분석, 검증, 보고서, 반복, 개선, 計画, 設計, 分析, 検証, 報告, 反復, 计划, 设计, 分析, 验证, 报告, 迭代, planificar, diseño, analizar, verificar, informe, planifier, conception, analyser, vérifier, rapport, planen, Entwurf, analysieren, überprüfen, Bericht, pianificare, progettazione, analizzare, verificare, rapporto Do NOT use for: simple one-line fixes, non-development tasks
Archived 2026-10-23 · Source updated 2026-05-20 · License: Apache-2.0. Check the original repository for the latest version.
SKILL.md instructions
# PDCA Skill
> Unified Skill for managing PDCA cycle. Supports the entire Plan → Design → Do → Check → Act flow.
## Arguments
| Argument | Description | Example |
|----------|-------------|---------|
| `plan [feature]` | Create Plan document | `/pdca plan user-auth` |
| `design [feature]` | Create Design document | `/pdca design user-auth` |
| `do [feature]` | Do phase guide | `/pdca do user-auth` |
| `analyze [feature]` | Run Gap analysis (Check) | `/pdca analyze user-auth` |
| `iterate [feature]` | Auto improvement (Act) | `/pdca iterate user-auth` |
| `report [feature]` | Generate completion report | `/pdca report user-auth` |
| `archive [feature]` | Archive PDCA documents | `/pdca archive user-auth` |
| `status` | Show current status | `/pdca status` |
| `next` | Guide to next phase | `/pdca next` |
## Action Details
### plan (Plan Phase)
1. Check if `docs/01-plan/features/{feature}.plan.md` exists
2. If not, create based on plan.template.md
3. Create Task: `[Plan] {feature}`
4. **Context Anchor Generation**: After generating Plan document, extract Context Anchor
(WHY/WHO/RISK/SUCCESS/SCOPE) from Executive Summary, Requirements, and Risk sections.
Write as `## Context Anchor` table between Executive Summary and Section 1.
| Dimension | Content |
|-----------|---------|
| WHY | Core problem being solved |
| WHO | Target users/stakeholders |
| RISK | Top 3 risks identified |
| SUCCESS | Measurable success criteria |
| SCOPE | In-scope / out-of-scope boundary |
### design (Design Phase)
1. Verify Plan document exists
2. Create `docs/02-design/features/{feature}.design.md`
3. Create Task: `[Design] {feature}` (blockedBy: Plan)
4. **Context Anchor Embed**: Copy Plan's `## Context Anchor` table to Design document top
(between header metadata and ## 1. Overview). If Plan has no Context Anchor, skip gracefully.
5. **Generate 3 Architecture Options**:
- Option A — Minimal Changes: Least modification, maximum reuse
- Option B — Clean Architecture: Best separation of concerns
- Option C — Pragmatic Balance: Good boundaries without over-engineering
6. Present comparison table with trade-offs (complexity, maintainability, effort, risk)
7. Ask user which option to select before proceeding
### do (Do Phase)
1. Verify Design document exists
2. Provide implementation guide
3. Create Task: `[Do] {feature}` (blockedBy: Design)
4. **Full Upstream Context Loading**: Read ALL upstream documents:
- PRD (docs/00-pm/{feature}.prd.md) — WHY context
- Plan (docs/01-plan/features/{feature}.plan.md) — Context Anchor, Success Criteria
- Design document — architecture decisions
5. **Decision Record Chain Display**:
```
Decision Record Chain
[PRD] Target: {segment} — {rationale}
[Plan] Architecture: {option} — {rationale}
[Design] Pattern: {approach} — {rationale}
```
6. **Display Context Anchor** from Design document header
7. **Parse --scope parameter**: If `--scope module-1` provided, filter implementation
guide to matching modules only from Design's Session Guide
8. Ask user to confirm scope before starting implementation
### analyze (Check Phase)
1. Call gap-detector agent
2. Compare Design vs implementation
3. Calculate Match Rate
4. Create Task: `[Check] {feature}`
5. After gap-detector analysis, also call `bkit_iterate` MCP tool for quantitative gap measurement:
- This tool compares design document vs actual implementation
- Returns matchRate percentage and specific gaps with file references
- If matchRate < 90%, recommend calling `/pdca iterate`
### iterate (Act Phase)
1. Check results (when matchRate < 90%)
2. Call pdca-iterator agent
3. Auto-fix and re-verify
4. Max 5 iterations
5. Call `bkit_iterate` MCP tool for automated gap analysis:
- Input: feature name, project directory, target match rate (default 90%)
- Tool returns gaps with specific file:line references
- Fix the identified gaps
- Call `bkit_iterate` again to re-check
- Repeat until matchRate >= 90% or max 5 iterations
- Server tracks iteration count internally
### qa (QA Phase)
When running QA, call `bkit_qa_run` MCP tool:
- Input: feature name, project directory, test levels [1,2,3,4,5]
- The tool auto-detects test framework (jest/vitest/playwright)
- L1: Unit tests, L2: API tests, L3-L5: E2E/browser tests (if playwright installed)
- Review results: passRate, coverage, defects
- If passRate < 95%, fix issues and re-run
### report (Completion Report)
1. Verify Check >= 90%
2. Call report-generator agent
3. Create completion report
4. Load ALL upstream documents for comprehensive reporting:
- PRD — compare original value proposition vs delivered value
- Plan — compare planned Success Criteria vs actual results
- Design — note architecture decisions and any deviations
- Analysis — include final Match Rate
5. Include Decision Record summary and Success Criteria final status
## PDCA Flow
```
[Plan] ✅ → [Design] ✅ → [Do] ✅ → [Check] 🔄 → [Act] ⏳ → [Report] 📋
↑_________|
(if < 90%)
```
## References
- `${extensionPath}/templates/plan.template.md`
- `${extensionPath}/templates/design.template.md`
- `${extensionPath}/templates/analysis.template.md`
- `${extensionPath}/templates/report.template.md`