Project Planning Skill

Version: 2.0 Last Updated: 2025-11-17 Status: Production Ready ✅

---

Purpose

This skill generates comprehensive, context-optimized planning documentation for web application projects. It structures work into manageable phases with built-in verification, ensuring projects can be built iteratively while maintaining clarity between sessions.

---

When to Use This Skill

Primary Use Cases

1. Starting a New Project (Most Common)

User: "I want to build a task management app with authentication"
Claude: "Let me use the project-planning skill to structure this properly"
→ Skill generates planning docs
→ Review and refine
→ Start building

2. Adding a Major Feature

User: "I need to add real-time collaboration to the existing app"
Claude: "Let me use project-planning to create phases for this feature"
→ Skill generates new phases
→ Append to existing IMPLEMENTATION_PHASES.md

3. Re-planning a Messy Project

User: "This project has gotten out of hand, let's restructure it"
Claude: "Let me use project-planning to reorganize the work"
→ Skill analyzes current state
→ Generates new phase structure

4. Validating Existing Phases

User: "Does Phase 3 look reasonable, or is it too big?"
Claude: "Let me use project-planning to validate this phase"
→ Skill checks sizing rules
→ Suggests improvements

---

What This Skill Generates

Core Output (Always)

• IMPLEMENTATION_PHASES.md - Structured phase breakdown with verification criteria
  • NEW in v1.1: Enhanced with file-level detail including:
    • File Maps - Shows which files to create/modify, their purpose, and dependencies
    • Data Flow Diagrams - Mermaid diagrams showing request/response flows
    • Critical Dependencies - Internal files, external packages, config, bindings
    • Gotchas & Known Issues - Security concerns, performance patterns, common pitfalls

Optional Outputs (Generated When Needed)

• DATABASE_SCHEMA.md - Tables, relationships, indexes, migrations
• API_ENDPOINTS.md - Routes, methods, auth, request/response schemas
• ARCHITECTURE.md - System design, data flow, service boundaries
• UI_COMPONENTS.md - Component hierarchy, state management, forms, phase-aligned installation strategy for shadcn/ui
• CRITICAL_WORKFLOWS.md - Complex setup steps, order-sensitive workflows, gotchas documentation
• INSTALLATION_COMMANDS.md - Copy-paste commands per phase (quick command reference)
• ENV_VARIABLES.md - All secrets, API keys, configuration guide (dev/prod setup)
• SESSION.md - Compact tracking template (<200 lines, references phases instead of duplicating)
• TESTING.md - Test strategy, E2E flows, integration tests
• AGENTS_CONFIG.md - AI agents, tools, workflows (for AI-enabled apps)
• INTEGRATION.md - Third-party services, webhooks, API integrations

Decision Logic: Skill asks if additional docs are needed, or auto-generates based on project complexity

• Database with >3 tables → Suggest DATABASE_SCHEMA.md
• API with >5 endpoints → Suggest API_ENDPOINTS.md
• Multiple services/workers → Suggest ARCHITECTURE.md
• AI-powered features → Suggest AGENTS_CONFIG.md

---

How It Works

1. Smart Defaults for Web Apps

The skill knows your preferred stack from CLAUDE.md:

• Frontend: Vite + React + Tailwind v4 + shadcn/ui
• Backend: Cloudflare Workers (with Static Assets for frontend)
• Database: D1 (SQL migrations)
• Storage: R2 (files), KV (config/cache)
• Auth: Clerk (JWT verification)
• Deployment: Wrangler CLI

It only asks clarifying questions when:

• Non-standard tech is mentioned
• Project complexity suggests alternatives
• User specifies different preferences

2. Hybrid Interaction

Template-driven with smart questions:

1. Analyzes project description
2. Identifies phase types needed (Infrastructure, Database, API, UI, Integration, Testing)
3. Asks targeted questions:
   • "Do you need social auth (Google, GitHub)?"
   • "Should users have roles/permissions?"
   • "Any real-time features (Durable Objects)?"
4. Generates docs with completed defaults and your specific choices

3. Phase Validation Rules

Every generated phase follows context-safe sizing:

• Max file scope: 5-8 files per phase
• Max dependencies: Shouldn't require deep knowledge of >2 other phases
• Time estimate: Includes implementation + verification + expected fixes
• Verification required: Every phase has checkbox criteria
• Exit criteria required: Clear "done" definition

If a phase violates these rules, skill auto-suggests splitting it.

---

Phase Types

The skill uses standardized templates for common web app patterns:

Infrastructure Phase

• Project scaffolding (Vite, Wrangler, dependencies)
• Build/deploy configuration
• Verification: Can deploy to Cloudflare, dev server runs

Database Phase

• Schema design (tables, relationships, indexes)
• Migration files
• Seed data for testing
• Verification: CRUD operations work, constraints enforced

API Phase

• Route definitions
• Middleware (auth, CORS, error handling)
• Request/response validation (Zod schemas)
• Verification: Endpoint tests pass (200, 401, 400, 500 cases)

UI Phase

• Component structure (shadcn/ui composition)
• Form validation (Zod + React Hook Form)
• State management (TanStack Query for server state, Zustand for client state)
• Verification: User flows work, forms validate, states update correctly

Integration Phase

• Third-party API setup (Clerk, Stripe, OpenAI, etc)
• Webhook handlers
• Configuration (environment variables, bindings)
• Verification: External service integration works, webhooks fire correctly

Testing Phase (Optional)

• E2E test setup
• Integration tests for critical flows
• Verification: Test suite passes, coverage meets threshold

---

File-Level Navigation (NEW in v1.1)

What Problem Does This Solve?

Before file maps: Claude needs to grep/glob through your codebase to understand where files are and what they do. This burns tokens and sometimes results in code being placed in the wrong files.

With file maps: Each phase includes a detailed map showing:

• Which files to create or modify
• What each file's purpose is
• Dependencies between files
• Security and performance considerations

Token Efficiency Gains

Example: "Implement task CRUD endpoints"

Enhanced Phase Structure

Each phase now includes (when applicable):

1. File Map

- `src/routes/tasks.ts` (estimated ~150 lines)
  - Purpose: CRUD endpoints for tasks
  - Key exports: GET, POST, PATCH, DELETE handlers
  - Dependencies: schemas.ts, auth middleware, D1 binding
  - Used by: Frontend task components

2. Data Flow Diagrams (Mermaid)

• Sequence diagrams for API calls
• Flowcharts for component logic
• Architecture diagrams for system components

3. Critical Dependencies

• Internal files (what imports what)
• External packages (with version hints)
• Configuration (env vars, Cloudflare bindings)

4. Gotchas & Known Issues

• Security patterns (ownership checks, auth)
• Performance considerations (pagination, caching)
• Framework quirks (Cloudflare Workers limits)

When File Maps Are Included

Always include for:

• API phases (prevents wrong endpoint placement)
• UI phases (shows component hierarchy)
• Integration phases (shows external service touchpoints)

Optional for:

• Infrastructure phases (scaffolding is self-evident)
• Database phases (schema files are self-documenting)

Example: Enhanced Phase

See references/example-enhanced-phase.md for complete before/after comparison with multiple Mermaid diagram examples.

---

New in v2.0: Enhanced Planning Outputs

1. Phase-Aligned UI Component Installation (shadcn/ui)

Problem Solved: Users were installing all shadcn components upfront, then forgetting which components each phase needed.

Solution: UI_COMPONENTS.md now includes phase-aligned installation strategy:

• Phase 1: Auth UI → Install button, input, label, card, sonner
• Phase 2: Forms → Install textarea, checkbox, select, switch
• Phase 3: Data Display → Install table, data-table, dialog, alert-dialog
• Quick reference commands (MVP install, full install, update all)
• Component usage by route mapping
• Design decisions documented (why sonner vs toast, etc.)

Time Saved: ~15-20 minutes per project (no more "which components do I need?")

2. Critical Workflows Documentation

Problem Solved: Complex, order-sensitive setup steps (like D1 binding order, better-auth factory patterns) were causing users to get stuck.

Solution: CRITICAL_WORKFLOWS.md documents:

• Step-by-step workflows with specific commands
• Why order matters (prevents "what breaks if done wrong")
• Common mistakes and how to recover
• Links to official docs and related skills

Example: D1 Database Binding workflow prevents the "cannot bind to non-existent Worker" error

Time Saved: ~30-60 minutes per complex workflow (no more trial-and-error)

3. Installation Commands Reference

Problem Solved: Users repeatedly asking "what was that command again?" between sessions.

Solution: INSTALLATION_COMMANDS.md provides:

• Copy-paste commands organized by phase
• Database commands (create, migrate, query)
• Deployment commands (deploy, set secrets, tail logs)
• Troubleshooting commands (clear cache, verify bindings)

Time Saved: ~5-10 minutes per session (instant command lookup)

4. Environment Variables Guide

Problem Solved: Users unsure which secrets/API keys needed, where to get them, how to configure.

Solution: ENV_VARIABLES.md includes:

• Development (.dev.vars) vs Production (wrangler secrets) setup
• Where to get each API key (with links)
• Security notes (what to commit, what to never commit)
• Setup checklists for local dev and production
• If secrets leaked recovery steps

Time Saved: ~20-30 minutes initial setup + prevents security mistakes

5. Compact SESSION.md Template

Problem Solved: SESSION.md was growing to 400+ lines by duplicating phase information from IMPLEMENTATION_PHASES.md.

Solution: New compact template (<200 lines):

• References IMPLEMENTATION_PHASES.md instead of duplicating
• Phase 0 + Phase 1 expanded, rest collapsed to 2-3 lines each
• Links to CRITICAL_WORKFLOWS.md prominently
• Critical Reminders section (what to check before starting)
• Known Risks section (high-risk phases identified upfront)

Token Saved: ~8-12k tokens per session (no duplicate phase info)

6. Pre-Planning Validation

Problem Solved: Users planning before prototyping new frameworks, missing critical workflows.

Solution: Skill now asks before generating docs:

1. "Have you built a prototype or POC?" → Suggests spike if no
2. "Any complex setup workflows discovered?" → Generates CRITICAL_WORKFLOWS.md if yes
3. "Tech stack familiarity?" → Adds learning curve buffer to estimates if learning

Time Saved: Prevents hours of backtracking from wrong assumptions

7. Post-Generation Validation Checklist

Problem Solved: Users starting Phase 1 without reviewing docs, environment setup, or critical workflows.

Solution: Skill outputs checklist after generating docs:

• Files: All docs reviewed, SESSION.md references correct files, critical workflows read
• Understanding: Phase 1 tasks understood, dependencies clear, risks identified
• Environment: GitHub repo, dev tools installed, accounts set up, CLI authenticated

Errors Prevented: "Missing IMPLEMENTATION_PHASES.md", "Forgot to set secrets", "Skipped critical workflow"

---

v2.0 Impact Summary

Production Tested: TanStack Cloudflare Starter project (revealed all these improvements)

---

Auto-Trigger Keywords

Claude Code will automatically discover this skill when you mention:

Project Initiation:

• "new project", "start a project", "create app", "build app", "scaffold project"
• "planning documentation", "project planning", "implementation phases"
• "structure project", "organize project", "plan features"

Planning Documents:

• "IMPLEMENTATION_PHASES.md", "DATABASE_SCHEMA.md", "API_ENDPOINTS.md"
• "ARCHITECTURE.md", "session handoff", "phase breakdown"
• "planning docs", "project spec", "requirements doc"

Phase Management:

• "phases", "phased approach", "implementation phases", "context-safe phases"
• "verification criteria", "exit criteria", "phase validation"
• "split phase", "merge phases", "adjust phases"

Stack Keywords:

• "Cloudflare Workers", "Vite + React", "D1 database", "full-stack app"
• "Clerk auth", "Tailwind v4", "shadcn/ui", "TanStack Query"

Use Cases:

• "major feature", "restructure project", "plan sprint", "roadmap"
• "session tracking", "context management", "git checkpoint"

AI/Integration Projects:

• "AI agents", "LLM features", "third-party integration", "webhooks"
• "real-time features", "Durable Objects", "websockets"

File-Level Navigation (NEW):

• "file map", "code navigation", "data flow", "mermaid diagram"
• "file dependencies", "gotchas", "known issues", "security patterns"
• "token efficiency", "which files", "where to start"

---

Invocation

Automatic (Recommended)

When you describe a new project, Claude Code will automatically suggest using this skill:

User: "I want to build a blog with comments and user profiles"
Claude: "Let me use the project-planning skill to structure this"

Manual

Explicitly request the skill:

"Use the project-planning skill to plan this project"
"Generate planning docs for this feature"
"Validate my existing phases with project-planning"

---

Example Workflow

User: "I want to build a URL shortener with analytics"

Skill:

Analyzing project requirements...

Project Type: Web App (CRUD + Analytics)
Suggested Stack: Cloudflare Workers + D1 + Vite + React

Questions:
1. Auth required? (public tool vs user accounts)
2. Custom short URLs or auto-generated?
3. Analytics detail level? (simple count vs detailed visitor data)
4. Admin dashboard needed?

User: "User accounts, auto-generated, detailed analytics, yes to dashboard"

Skill Generates:

• IMPLEMENTATION_PHASES.md (6 phases)
• DATABASE_SCHEMA.md (users, urls, analytics_events tables)
• API_ENDPOINTS.md (auth, URL CRUD, redirect, analytics)

Output Preview:

# Implementation Phases: URL Shortener

## Phase 1: Project Setup (2 hours)
**Type**: Infrastructure
**Files**: package.json, wrangler.jsonc, vite.config.ts, src/index.ts

**Tasks**:
- [x] Scaffold Cloudflare Worker with Vite
- [x] Configure Tailwind v4 + shadcn/ui
- [x] Setup D1 database binding
- [x] Test deployment

**Verification**:
- [ ] `npm run dev` starts without errors
- [ ] Can deploy to Cloudflare
- [ ] Worker serves React app

**Exit Criteria**: Working dev environment and successful deployment

---

## Phase 2: Database Schema (2-3 hours)
**Type**: Database
**Files**: migrations/0001_initial.sql, src/db/schema.ts

[... and so on for each phase ...]

---

What This Skill Does NOT Do

Not a code generator: Generates planning docs, not implementation code Not a session tracker: Doesn't update SESSION.md (that's Claude's job) Not an architect: Doesn't make technical decisions (Claude does that with you) Not mandatory: Optional tool - use when beneficial

---

Expanding the Skill

Easy to add new document templates:

1. Create template in templates/ directory
2. Add generation logic to SKILL.md
3. Update this README

Future possibilities:

• phase-split: Break oversized phase into sub-phases
• session-init: Generate SESSION.md from IMPLEMENTATION_PHASES.md
• More phase types (Migration, Optimization, Security)

---

Files in This Skill

project-planning/
├── README.md                          # This file
├── SKILL.md                           # Main skill logic (planning assistant)
├── templates/
│   ├── IMPLEMENTATION_PHASES.md       # Phase breakdown template (enhanced with file maps)
│   ├── DATABASE_SCHEMA.md             # Database design template
│   ├── API_ENDPOINTS.md               # API routes template
│   ├── ARCHITECTURE.md                # System design template
│   ├── UI_COMPONENTS.md               # Component structure template
│   ├── TESTING.md                     # Test strategy template
│   ├── AGENTS_CONFIG.md               # AI agents template
│   └── INTEGRATION.md                 # Third-party integrations template
└── references/
    ├── example-enhanced-phase.md      # NEW: File maps, Mermaid diagrams, before/after
    └── example-outputs/
        ├── simple-web-app.md          # Basic CRUD example
        ├── auth-web-app.md            # Authentication example
        └── ai-web-app.md              # AI-powered app example

---

Tips for Best Results

Be specific in project description: "Task manager with tags and due dates" is better than "to-do app"

Mention non-standard requirements early: "Need to support 10k concurrent users" or "Must integrate with legacy SOAP API"

Review generated phases before starting: Easier to refine planning docs than mid-implementation

Use examples for reference: Check references/example-outputs/ directory for similar projects

Iterate if needed: Regenerate or refine phases if initial output doesn't feel right

---

Integration with Session Workflow

After this skill generates IMPLEMENTATION_PHASES.md:

1. Review and refine phases
2. Create SESSION.md (manually or with future session-init tool)
3. Start Phase 1
4. Update SESSION.md as you progress
5. Create git checkpoints at phase boundaries

See main CLAUDE.md for full session handoff protocol.

---

Questions?

This skill is designed to evolve based on usage. If something doesn't work as expected or you'd like additional capabilities, refine SKILL.md and templates directly.