Workflow Generation Skill
Purpose: Provide deep knowledge about organizing software development commands into phased workflows for parallel execution and clear dependency management.
Core Concepts
The Phase-Based Execution Model
Philosophy: Software development commands should be organized into distinct phases that can be:
- Executed in separate terminal sessions (fresh context)
- Run in parallel where dependencies allow
- Used as clear checkpoints in the development process
Why Phases Matter:
- Avoid context contamination - Each terminal starts fresh
- Enable parallel work - Multiple developers or instances
- Create natural checkpoints - Complete one phase before next
- Facilitate rollback - Easy to return to phase boundaries
Phase Organization Patterns
Standard Dev Lifecycle Phases
Phase 1: Foundation (Always first, no dependencies)
- Purpose: Set up project structure and environment
- Typical commands:
- Project initialization
- Directory structure creation
- Environment detection/validation
- Repository initialization
- Secret management setup
- Tool installation verification
Phase 2: Planning (Requires: Foundation complete)
- Purpose: Define what to build before building it
- Typical commands:
- Requirements gathering
- Spec creation
- Architecture design
- Decision documentation (ADRs)
- Roadmap creation
- Database schema design
Phase 3: Implementation (Requires: Planning + Foundation)
- Purpose: Build the actual features
- Typical commands:
- Task layering
- Component creation
- Feature development
- Integration setup
- Data model implementation
- Sub-phases:
- L0: Infrastructure (databases, APIs, auth)
- L1: Core components (shared libraries)
- L2: Features (business logic)
- L3: Integration (wiring everything together)
Phase 4: Quality (Can run parallel with late Implementation)
- Purpose: Validate what was built
- Typical commands:
- Test suite generation
- Security scanning
- Code validation
- Performance analysis
- Accessibility testing
Phase 5: Deployment (Requires: Quality passed)
- Purpose: Ship to production
- Typical commands:
- Platform detection
- Pre-flight checks
- CI/CD setup
- Deployment execution
- Health validation
- Monitoring setup
Phase 6: Iteration (Ongoing, throughout process)
- Purpose: Improve and refine
- Typical commands:
- Feature enhancement
- Code refactoring
- Spec synchronization
- Feedback integration
Command Classification Patterns
How to Categorize Commands
Foundation Indicators:
- Keywords:
init, setup, detect, validate, env, structure
- Actions: Creates directories, validates tools, sets up environment
- Examples:
/foundation:init-structure, /foundation:env-check
Planning Indicators:
- Keywords:
wizard, spec, architecture, decide, roadmap, plan
- Actions: Creates documentation, designs systems, documents decisions
- Examples:
/planning:wizard, /planning:architecture
Implementation Indicators:
- Keywords:
add, create, integrate, build, generate
- Actions: Creates code, components, features
- Examples:
/nextjs-frontend:add-component, /fastapi-backend:add-endpoint
Quality Indicators:
- Keywords:
test, validate, security, performance, quality
- Actions: Runs tests, scans for issues, validates code
- Examples:
/testing:test, /quality:validate-code
Deployment Indicators:
- Keywords:
deploy, prepare, cicd, validate, monitor
- Actions: Deploys to platforms, sets up pipelines, validates deployment
- Examples:
/deployment:deploy, /deployment:prepare
Iteration Indicators:
- Keywords:
enhance, refactor, adjust, sync, improve
- Actions: Improves existing code, syncs documentation
- Examples:
/iterate:enhance, /iterate:refactor
Phase Dependency Rules
Strict Dependencies (MUST follow order)
Foundation → Planning → Implementation → Quality → Deployment
- Foundation has no dependencies (always first)
- Planning requires Foundation complete
- Implementation requires Planning + Foundation
- Quality requires Implementation (at least partially)
- Deployment requires Quality passed
- Iteration can happen throughout (but often after initial implementation)
Parallel Execution Opportunities
Within Foundation:
- Environment checks can run parallel with structure validation
- Secret management can run parallel with repository init
Within Implementation:
- Frontend and backend can be built in parallel
- Different features can be built simultaneously (if independent)
Quality and Late Implementation:
- Tests can be written while final features are being implemented
- Security scanning can run on completed components
Project Context Reading Patterns
What to Read (In Order)
-
.claude/project.json (MUST READ FIRST)
- Tech stack information
- Framework versions
- Architecture pattern
- Deployment targets
-
docs/architecture/ (If exists)
- System design
- Component diagrams
- Data flow
- Integration points
-
docs/adr/ (If exists)
- Architecture decisions
- Decision rationale
- Trade-offs made
- Migration plans
-
specs/ (If exists)
- Feature specifications
- Requirements
- Task breakdowns
-
Current file state (File existence checks)
- What's already implemented
- What's missing
- Completion status
How to Use Context
From project.json:
- Extract tech stack (determines which implementation commands to include)
- Understand architecture pattern (monolithic vs microservices affects phasing)
- Identify deployment targets (determines deployment commands)
From ADRs:
- Understand WHY decisions were made
- Include relevant context in workflow
- Respect architectural patterns chosen
From architecture docs:
- Understand system design
- Include in workflow overview
- Guide implementation phase organization
From specs:
- Know what features exist
- Understand current state
- Determine what still needs building
From file existence:
- Mark commands as ✅ (done) if files exist
- Mark as □ (todo) if files missing
- Show progress through workflow
Workflow Document Structure
Recommended Sections
-
Project Overview
- Tech stack summary (from project.json)
- Architecture pattern (from ADRs)
- Current status (from file checks)
-
Phased Commands
- One section per phase
- Clear phase boundaries
- Commands grouped logically
- Dependencies noted
-
Progress Indicators
- ✅ = Already done (auto-detected)
- □ = Still to do
- 🔄 = In progress (partial implementation)
-
Project-Specific Context
- Relevant ADR summaries
- Architecture highlights
- Business metrics (if applicable)
-
Next Steps
- What to run next
- What's blocking progress
- How to regenerate workflow
Examples from Real Projects
Example: AI Education Platform (Monolithic MVP)
Context from ADRs:
- ADR 0006: Start monolithic, migrate to microservices later
- Trigger: >5,000 students or >30 sec AI response times
Phase Organization:
Phase 1: Foundation
- /foundation:init-structure
- /foundation:github-init
- /foundation:doppler-setup
Phase 2: Planning
- /planning:wizard
- /planning:architecture
- /planning:decide
Phase 3: Implementation (Monolithic)
- /fastapi-backend:init (single service)
- /nextjs-frontend:init
- /supabase:init
- Implement study partner as embedded library (NOT microservice)
Phase 4-6: Standard (Quality, Deployment, Iteration)
Key Insight: Architecture decision (monolithic) directly affects implementation phase.
Integration with Dan's Composition Pattern
How This Skill Fits
Slash Command (/foundation:generate-workflow)
↓
Loads THIS SKILL (knowledge about phasing)
↓
Runs Python script (bulk Airtable query)
↓
Uses skill knowledge to organize commands
↓
Outputs workflow markdown
This skill provides:
- Phase organization knowledge
- Command classification patterns
- Dependency rules
- Project context patterns
The slash command applies:
- Reads project context
- Applies skill patterns
- Makes intelligent decisions
- Generates final workflow
The Python script provides:
- Raw Airtable data
- Command validation
- Bulk operations
Best Practices
Do's ✅
- Always read project.json first - It drives everything else
- Respect architecture decisions - ADRs guide phasing
- Validate commands - Check Airtable vs filesystem
- Use file existence - Auto-detect completion status
- Clear phase boundaries - Easy to execute in separate terminals
- Include project context - ADRs, architecture, business metrics
Don'ts ❌
- Don't hardcode phases in Python - Intelligence belongs in slash command
- Don't ignore ADRs - They contain critical context
- Don't assume command availability - Always validate
- Don't mix phase boundaries - Keep them distinct
- Don't generate static workflows - Make them project-aware
Validation Checklist
When generating a workflow, ensure:
Regeneration Pattern
Always include regeneration command in output:
# To regenerate this workflow with latest Airtable data:
/foundation:generate-workflow "Your Tech Stack Name"
This allows users to:
- Update workflow when Airtable changes
- Refresh completion status
- Incorporate new commands
- Re-validate against filesystem
This skill enables intelligent, project-aware workflow generation that respects architecture decisions and enables parallel execution.