Claude-Code-Workflow/docs/skills/claude-workflow.md

Claude Skills - Workflow

One-Liner

Workflow Skills is a task orchestration and execution pipeline system — Complete automation from planning to execution to verification.

Pain Points Solved

Pain Point	Current State	Claude_dms3 Solution
Manual task breakdown	Manual decomposition, easy to miss	Automated task generation and dependency management
Scattered execution state	Tools independent, state not unified	Unified session management, TodoWrite tracking
Difficult interruption recovery	Hard to resume after interruption	Session persistence, checkpoint recovery
Missing quality verification	No quality check after completion	Built-in quality gates, verification reports

Skills List

Skill	Function	Trigger
workflow-plan	Unified planning skill (4-stage workflow)	/workflow:plan
workflow-execute	Agent-coordinated execution	/workflow:execute
workflow-lite-plan	Lightweight quick planning	/workflow:lite-plan
workflow-multi-cli-plan	Multi-CLI collaborative planning	/workflow:multi-cli-plan
workflow-tdd	TDD workflow	/workflow:tdd
workflow-test-fix	Test-fix workflow	/workflow:test-fix
workflow-skill-designer	Skill design workflow	/workflow:skill-designer
workflow-wave-plan	Wave batch planning	/workflow:wave-plan

Skills Details

workflow-plan

One-Liner: Unified planning skill — 4-stage workflow, plan verification, interactive re-planning

Trigger:

/workflow:plan <task-description>
/workflow:plan-verify --session <session-id>
/workflow:replan --session <session-id> [task-id] "requirements"

Features:

• 4-stage workflow: Session discovery → Context collection → Conflict resolution → Task generation
• Plan verification (Phase 5): Read-only verification + quality gates
• Interactive re-planning (Phase 6): Boundary clarification → Impact analysis → Backup → Apply → Verify

Mode Detection:

// Skill trigger determines mode
skillName === 'workflow:plan-verify' → 'verify'
skillName === 'workflow:replan' → 'replan'
default → 'plan'

Core Rules:

1. Pure coordinator: SKILL.md only routes and coordinates, execution details in phase files
2. Progressive phase loading: Read phase documents only when phase is about to execute
3. Multi-mode routing: Single skill handles plan/verify/replan via mode detection
4. Task append model: Subcommand tasks are appended, executed sequentially, then collapsed
5. Auto-continue: Automatically execute next pending phase after each phase completes
6. Accumulated state: planning-notes.md carries context across phases for N+1 decisions

Plan Mode Data Flow:

User Input (task description)
    ↓
[Convert to structured format]
    ↓ Structured description:
    ↓   GOAL: [goal]
    ↓   SCOPE: [scope]
    ↓   CONTEXT: [context]
    ↓
Phase 1: session:start --auto "structured-description"
    ↓ Output: sessionId
    ↓ Write: planning-notes.md (user intent section)
    ↓
Phase 2: context-gather --session sessionId "structured-description"
    ↓ Input: sessionId + structured description
    ↓ Output: contextPath + conflictRisk
    ↓ Update: planning-notes.md
    ↓
Phase 3: conflict-resolution [condition: conflictRisk ≥ medium]
    ↓ Input: sessionId + contextPath + conflictRisk
    ↓ Output: Modified brainstorm artifacts
    ↓ Skip if conflictRisk is none/low → go directly to Phase 4
    ↓
Phase 4: task-generate-agent --session sessionId
    ↓ Input: sessionId + planning-notes.md + context-package.json
    ↓ Output: IMPL_PLAN.md, task JSONs, TODO_LIST.md
    ↓
Plan Confirmation (User Decision Gate):
    ├─ "Verify plan quality" (recommended) → Phase 5
    ├─ "Start execution" → Skill(skill="workflow-execute")
    └─ "Review status only" → Inline show session status

TodoWrite Mode:

• Task append (during phase execution): Subtasks appended to coordinator's TodoWrite
• Task collapse (after subtask completion): Remove detailed subtasks, collapse to phase summary
• Continuous execution: Auto-proceed to next pending phase after completion

---

workflow-execute

One-Liner: Agent-coordinated execution — Systematic task discovery, agent coordination, and state tracking

Trigger:

/workflow:execute
/workflow:execute --resume-session="WFS-auth"
/workflow:execute --yes
/workflow:execute -y --with-commit

Features:

• Session discovery: Identify and select active workflow sessions
• Execution strategy resolution: Extract execution model from IMPL_PLAN.md
• TodoWrite progress tracking: Real-time progress tracking throughout workflow execution
• Agent orchestration: Coordinate specialized agents with full context
• Autonomous completion: Execute all tasks until workflow completes or reaches blocking state

Auto Mode Defaults (--yes or -y):

• Session selection: Auto-select first (latest) active session
• Completion selection: Auto-complete session (run /workflow:session:complete --yes)

Execution Process:

Phase 1: Discovery
   ├─ Count active sessions
   └─ Decision:
      ├─ count=0 → Error: No active sessions
      ├─ count=1 → Auto-select session → Phase 2
      └─ count>1 → AskUserQuestion (max 4 options) → Phase 2

Phase 2: Plan Document Verification
   ├─ Check IMPL_PLAN.md exists
   ├─ Check TODO_LIST.md exists
   └─ Verify .task/ contains IMPL-*.json files

Phase 3: TodoWrite Generation
   ├─ Update session status to "active"
   ├─ Parse TODO_LIST.md for task status
   ├─ Generate TodoWrite for entire workflow
   └─ Prepare session context paths

Phase 4: Execution Strategy Selection & Task Execution
   ├─ Step 4A: Parse execution strategy from IMPL_PLAN.md
   └─ Step 4B: Lazy load execution tasks
      └─ Loop:
         ├─ Get next in_progress task from TodoWrite
         ├─ Lazy load task JSON
         ├─ Start agent with task context
         ├─ Mark task complete
         ├─ [with-commit] Commit changes based on summary
         └─ Advance to next task

Phase 5: Completion
   ├─ Update task status
   ├─ Generate summary
   └─ AskUserQuestion: Select next action

Execution Models:

Model	Condition	Mode
Sequential	IMPL_PLAN specifies or no clear parallelization guidance	Execute one by one
Parallel	IMPL_PLAN specifies parallelization opportunities	Execute independent task groups in parallel
Phased	IMPL_PLAN specifies phase breakdown	Execute by phase, respect phase boundaries
Intelligent Fallback	IMPL_PLAN lacks execution strategy details	Analyze task structure, apply intelligent defaults

Lazy Loading Strategy:

• TODO_LIST.md: Phase 3 read (task metadata, status, dependencies)
• IMPL_PLAN.md: Phase 2 check existence, Phase 4A parse execution strategy
• Task JSONs: Lazy load — read only when task is about to execute

Auto Commit Mode (--with-commit):

# 1. Read summary from .summaries/{task-id}-summary.md
# 2. Extract files from "Files Modified" section
# 3. Commit: git add <files> && git commit -m "{type}: {title} - {summary}"

---

workflow-lite-plan

One-Liner: Lightweight quick planning — Quick planning and execution for super simple tasks

Trigger:

/workflow:lite-plan <simple-task>

Features:

• For Level 1 (Lite-Lite-Lite) workflow
• Minimal planning overhead for super simple quick tasks
• Direct text input, no complex analysis

Use Cases:

• Small bug fixes
• Simple documentation updates
• Configuration adjustments
• Single function modifications

---

workflow-multi-cli-plan

One-Liner: Multi-CLI collaborative planning — Analysis and planning with multiple CLI tools collaborating

Trigger:

/workflow:multi-cli-plan <task>

Features:

• Call multiple CLI tools (Gemini, Codex, Claude) for parallel analysis
• Synthesize input from multiple perspectives
• Generate comprehensive plan

Use Cases:

• Tasks requiring multi-perspective analysis
• Complex architecture decisions
• Cross-domain problems

---

workflow-tdd

One-Liner: TDD workflow — Test-driven development process

Trigger:

/workflow:tdd <feature-description>

Features:

• Write tests first
• Implement functionality
• Run tests to verify
• Loop until passing

Pipeline:

Plan Tests → Write Tests → [Fail] → Implement Feature → [Pass] → Complete
                    ↑___________|

---

workflow-test-fix

One-Liner: Test-fix workflow — Diagnosis and fixing of failing tests

Trigger:

/workflow:test-fix <failing-tests>

Features:

• Diagnose test failure causes
• Fix code or tests
• Verify fixes
• Loop until passing

Pipeline:

Diagnose Failure → Identify Root Cause → [Code Issue] → Fix Code → Verify
                          ↑___________|

---

workflow-skill-designer

One-Liner: Skill design workflow — Create new Claude Code Skills

Trigger:

/workflow:skill-designer <skill-idea>

Features:

• Requirement discovery
• Structure generation
• Phase/action generation
• Specification and template generation
• Verification and documentation

Output Structure:

.claude/skills/{skill-name}/
├── SKILL.md              # Entry file
├── phases/
│   ├── orchestrator.md   # Orchestrator
│   └── actions/          # Action files
├── specs/                # Specification documents
├── templates/            # Template files
└── README.md

---

workflow-wave-plan

One-Liner: Wave batch planning — Parallel processing planning for batch issues

Trigger:

/workflow:wave-plan <issue-list>

Features:

• Batch issue analysis
• Parallelization opportunity identification
• Wave scheduling (batch by batch)
• Execution queue generation

Wave Model:

Wave 1: Issue 1-5 → Parallel planning → Parallel execution
Wave 2: Issue 6-10 → Parallel planning → Parallel execution
...

Related Commands

• Claude Commands - Workflow
• Claude Skills - Team Collaboration

Best Practices

1. Choose the right workflow:

   • Super simple tasks → workflow-lite-plan
   • Complex features → workflow-plan → workflow-execute
   • TDD development → workflow-tdd
   • Test fixes → workflow-test-fix
   • Skill creation → workflow-skill-designer

2. Leverage auto mode: Use --yes or -y to skip interactive confirmations, use defaults

3. Session management: All workflows support session persistence, can resume after interruption

4. TodoWrite tracking: Use TodoWrite to view workflow execution progress in real-time

5. Lazy Loading: Task JSONs are lazy loaded, read only during execution, improving performance