workflow-guide

# Workflow Selection Guide (Meta Hub) > **Purpose**: Auto-analyze project state → recommend 1–2 optimal skills > > **Core principle**: This skill **does not write implementation code**. It only performs **situation diagnosis → skill recommendation → user confirmation**. --- ## Absolute Prohibitions 1. Do not write code directly — guide the workflow only. 2. Do not list all skills — recommend **1–2 only**, matched to the situation. 3. Do not start with requirements questions — **auto-diagnose** project state first. --- ## Actions to Execute Immediately on Skill Activation ### Stage 1: Auto-diagnose project state (Silent Analysis) ```bash # 1. Check for task file ls TASKS.md 2>/dev/null || ls docs/planning/06-tasks.md 2>/dev/null # 2. Check codebase (actual code files in src/·app/·lib/) SOURCE_CODE=$(find src/ app/ lib/ -maxdepth 3 -type f \( -name "*.ts" -o -name "*.js" -o -name "*.py" \) 2>/dev/null | head -1) echo "source_code=${SOURCE_CODE:+yes}" # 3. Check for interrupted work (state file + incomplete tasks) for STATE_PATH in ".claude/orchestrate-state.json" ".claude/orchestrate/orchestrate-state.json"; do if [ -f "$STATE_PATH" ] && grep -qE '"status"[[:space:]]*:[[:space:]]*"(in_progress|pending)"' "$STATE_PATH" 2>/dev/null; then echo "state_file=$STATE_PATH incomplete_in_state=1" fi done # 4. Check for git merge conflicts CONFLICT_FILES=$(git diff --name-only --diff-filter=U 2>/dev/null | wc -l | tr -d ' ') echo "conflicts=$CONFLICT_FILES" # 5. Agent/task counts AGENT_COUNT=$(ls .claude/agents/*.md 2>/dev/null | wc -l | tr -d ' ') TASK_COUNT=$(grep -cE '^\s*[-*]\s*\[|^#{1,6}\s+\[' TASKS.md 2>/dev/null || echo 0) INCOMPLETE_COUNT=$(grep -cE '^\s*[-*]\s*\[\s*\]|^#{1,6}\s+\[\s*\]' TASKS.md 2>/dev/null || echo 0) GOVERNANCE_DONE=$(ls management/project-plan.md 2>/dev/null && echo "yes" || echo "no") echo "agents=$AGENT_COUNT tasks=$TASK_COUNT incomplete=$INCOMPLETE_COUNT governance=$GOVERNANCE_DONE" # 6. cmux 감지 CMUX_AVAILABLE=$(cmux ping 2>/dev/null && echo "yes" || echo "no") echo "cmux=$CMUX_AVAILABLE" ``` ### Stage 2: Decision Algorithm (rules + experience hybrid) > workflow-guide does **current state → recommend 1 skill**. > When experience data exists, memento's Smart Router can override the rule-based recommendation. ``` ① Recovery: state file incomplete OR merge conflicts → /recover ② No tasks: no TASKS.md → /tasks-init (or /tasks-migrate if legacy exists) ③ Maintenance: source_code exists + bug/fix request → /maintenance ③-b Maintenance (iterative): source_code exists + incomplete>0 → /agile iterate ④ New implementation: incomplete>0 + TASK<30 → /agile auto ⑤ Large-scale (30-50): incomplete>=30 → /team-orchestrate --mode=auto (or --mode=team if Agent Teams) ⑤-b Very large (50+): incomplete>=50 → /team-orchestrate --mode=thin ⑥ Done: all_tasks_completed → /audit ``` **Memento Experience Override** (when `.claude/memento/experience.jsonl` exists with 5+ entries): After computing the rule-based recommendation, check if memento's experience-weighted router suggests a different skill with confidence > 0.75. If so, use the experience-based recommendation and note the override reason. See `/memento route` for the full algorithm. This ensures backward compatibility — without experience data, routing is purely rule-based. **cmux 오버라이드** (CMUX_AVAILABLE=yes일 때 ⑤ 이후 적용): | 기본 추천 | cmux 대체 | 이유 | |-----------|-----------|------| | `/team-orchestrate` | `/cmux-orchestrate` | 물리적 계층 + 이기종 AI 팀 | | `/multi-ai-run` | `/cmux-ai-run` | 창 분할 진짜 병렬 실행 | | `/multi-ai-review` | `/cmux-ai-review` | 패널 나란히 동시 리뷰 | cmux 감지 시 추천 메시지에 cmux 변형 스킬을 ⭐ 추천으로 올리고, 기존 스킬은 대안으로 제시. Prerequisite checks each skill performs on its own: - `/team-orchestrate` → checks TASKS.md format and Agent Teams installation - `/agile` → checks TASKS.md existence/format, recommends team-orchestrate for ≥30 tasks - `/audit` → checks for planning documents, guides to /governance-setup if missing - `/governance-setup` → guides to /tasks-init or /tasks-migrate based on TASKS.md state ### Stage 3: Tailored Recommendation (AskUserQuestion) Display the diagnosis result with a starred recommendation for user confirmation: ```json { "questions": [{ "question": "Project state has been analyzed. Choose the next step:", "options": [ { "label": "⭐ [Recommended] {diagnosed skill}", "description": "{reason for fit}" }, { "label": "{alternative 1}", "description": "{description}" } ] }] } ``` --- ## Standalone Skill Catalog (25 skills) | Skill | Trigger | Role | |-------|---------|------| | **`/workflow`** | "what should I do?" | Meta hub — skill routing | | **`/governance-setup`** | `/governance-setup` | Governance + Mini-PRD planning | | **`/tasks-init`** | `/tasks-init` | TASKS.md scaffolding | | **`/tasks-migrate`** | `/tasks-migrate` | Consolidate legacy tasks | | **`/agile`** | `/agile auto` | Layer-based sprint execution | | **`/team-orchestrate`** | `/team-orchestrate`, `/auto-orchestrate` | Unified orchestration — auto/team/thin modes | | **`/cmux-orchestrate`** | `/cmux-orchestrate` | cmux 물리적 3-Level 팀 (Claude/Gemini/Codex 혼합) | | **`/multi-ai-run`** | `/multi-ai-run` | Role-based model routing | | **`/cmux-ai-run`** | `/cmux-ai-run` | cmux 창 분할 병렬 태스크 실행 | | **`/checkpoint`** | "review this" | Two-stage review on task completion | | **`/security-review`** | `/security-review` | OWASP Top 10 security scan | | **`/audit`** | `/audit` | Pre-deployment comprehensive audit | | **`/multi-ai-review`** | `/multi-ai-review` | 3-AI consensus review | | **`/cmux-ai-review`** | `/cmux-ai-review` | cmux 창 분할 병렬 3-Stage 리뷰 | | **`/recover`** | "work was interrupted" | Work recovery hub | | **`/impact`** | `/impact <file>` | Change impact analysis | | **`/deps`** | `/deps` | Dependency graph | | **`/coverage`** | `/coverage` | Test coverage | | **`/architecture`** | `/architecture` | Architecture map | | **`/maintenance`** | "fix this bug" | ITIL 5-stage production maintenance orchestrator | | **`/compress`** | "compress context" | Long context optimization | | **`/statusline`** | auto-activated | Progress status bar display | | **`/changelog`** | `/changelog` | Change history query | | **`/cmux`** | `/cmux` | cmux 터미널 멀티플렉서 제어 | | **`/memento`** | "skill health", "which skill" | Skill ecosystem intelligence — experience logging, smart routing, health dashboard | --- ## Skill Selection by Task Scale | Task count | Recommended skill | Mode | Prerequisite skills | |------------|-------------------|------|---------------------| | **1–30** | `/agile auto` | — | — | | **30–50** | `/team-orchestrate` | `--mode=auto` or `--mode=team` | `/governance-setup` (optional) | | **50–200** | `/team-orchestrate` | `--mode=thin` | `/governance-setup` | | **200+** | `/team-orchestrate --mode=thin --phase N` | split by phase | `/governance-setup` | --- ## Natural Language → Skill Quick Map (GSD 스타일 확장) ``` "I don't know where to start" → /workflow "I have a spec, start coding" → /agile auto "Fix this feature" → /agile iterate "Review the code" → /checkpoint "Do a deep review" → /multi-ai-review (cmux: /cmux-ai-review) "Run a security check" → /security-review "Run a quality check" → /audit "Work was interrupted" → /recover "This is a large project" → /governance-setup "Fix this bug" → /maintenance "Fix this in production" → /maintenance "Run with an agent team" → /team-orchestrate (cmux: /cmux-orchestrate) "Use multiple AIs in parallel" → /cmux-ai-run (cmux 감지 시) / /multi-ai-run "Show execution status" → /whitebox status "Compress the context" → /compress "Which skill works best for this?" → /memento route "Show skill performance" → /memento health "Why does this skill keep failing?" → /memento reflect # GSD 스타일 자연어 라우팅 "이거 리팩토링해줘" → /agile auto (복잡) 또는 /agile iterate (단순) "이 기능 추가해줘" → /discuss → /agile auto "뭘 어떻게 해야 할지 모르겠어" → /discuss "결정해야 할 게 있어" → /discuss "요구사항 분석해줘" → /si-planning "기능정의서 만들어줘" → /si-planning "화면정의서 만들어줘" → /si-planning "SI 기획 시작" → /si-planning --domain=? "요구사항 변경" → /si-planning --change "누락 검사" → /si-planning --gap "고객 대시보드" → /si-planning --dashboard "이거 검증해줘" → /audit (goal-backward 포함) "코드 확인해줘" → /checkpoint "이거 맞게 했는지 봐줘" → /audit "자동으로 다 해줘" → /team-orchestrate --mode=autonomous "알아서 해줘" → /agile auto ``` ### 자동 라우팅 모드 `--auto` 플래그 추가 시 사용자 확인 없이 바로 실행: ```bash # 기존 /workflow → 진단 → 추천 → 사용자 확인 → 실행 # --auto 모드 /workflow --auto → 진단 → 바로 실행 ``` --- ## Reference Documents For detailed algorithms, skill integration matrix, and quality gates, see: - `references/decision-algorithm.md` — Detailed decision algorithm + scenario validation - `references/skill-integrations.md` — Skill-to-skill integrations + failure recovery paths --- **Last Updated**: 2026-03-18 (v5.1.0 — cmux auto-detection + cmux skill routing added)