claude-project-planner

Claude Project Planner

AI-powered project planning toolkit that transforms ideas into comprehensive, buildable specifications. Generate complete project plans including architecture documents, sprint plans, cost analyses, and implementation roadmaps—all backed by real-time research.

✨ New in v1.4.5: Simpler, faster, cheaper! Architecture optimization removes 8.6K lines of unnecessary code, adds intelligent cost optimization (29-57% search cost savings), and achieves 50-85% faster responses through optimized caching. With Claude MAX subscription, total cost per run is just $0.15-0.30 (Perplexity searches only).

---

🚀 Quick Start

Installation

# Install as Claude Code plugin (recommended)
claude plugin install flight505/claude-project-planner

# Restart Claude Code when prompted

Prerequisites

• Python 3.10-3.12
• ANTHROPIC_API_KEY (required)
• OPENROUTER_API_KEY (recommended, for research)
• GEMINI_API_KEY (optional, for Deep Research)

# Configure API keys
export ANTHROPIC_API_KEY='your_key'
export OPENROUTER_API_KEY='your_key'  # Recommended for research

Usage

# Start comprehensive planning with interactive setup
/full-plan my-project

# You'll see 8 question groups covering ALL features:
# 1. AI Provider (Gemini vs Perplexity)
# 2. Research Depth (balanced/quick/comprehensive)
# 3. Parallelization (fast vs sequential)
# 4. Interactive Approval (pause after each phase)
# 5. Core Phases (Market Research, Architecture, Implementation)
# 6. Optional Phases (Feasibility, Go-to-Market, Review)
# 7. Quality Checks (multi-model validation, diagrams)
# 8. Output Formats (PDF, PowerPoint)

That's it! No flags to remember—discover all features through the interactive UI.

---

✨ What's New in v1.4.5

🚀 Architecture Optimization & Performance Improvements

Simpler, faster, cheaper! v1.4.5 removes unnecessary complexity and adds intelligent cost optimization:

✅ Removed Unnecessary Systems (~8.6K lines)

• Eliminated Gemini budget tracking (was based on incorrect subscription assumption)
• Removed redundant streaming wrapper (Claude Agent SDK already provides streaming)
• Cleaner codebase with same functionality

⚡ Smart Cost Optimization

• Dynamic Context Sizing: Automatically selects low/medium/high context based on query complexity
  • Simple queries: $6/1K searches (57% savings)
  • Standard queries: $10/1K searches (29% savings)
  • Complex queries: $14/1K searches (maintains depth)
• Optimized Tool Loading: Restricts to essential tools → 30-40% smaller cache → 50-85% faster responses after first request
• Cost Tracking: Research responses now include context_size metadata for transparency

📊 Impact:

• Code: 18% reduction (simpler architecture)
• Search costs: $0.15-0.30/run (vs $0.30-0.50/run)
• Response speed: 50-85% faster (after cache warm-up)
• Quality: Unchanged (same citations and research depth)

💰 Cost with Claude MAX: If you have Claude MAX subscription ($60/month), Claude API costs are $0 (included). Total cost per run: ~$0.15-0.30 (just Perplexity searches).

See: REFACTOR_SUMMARY.md and ARCHITECTURE_AUDIT_RECOMMENDATIONS.md for comprehensive analysis.

---

🎯 How It Works

The Simple View (ELI5)

Think of Claude Project Planner as a smart assistant that turns your idea into a complete project blueprint. Instead of manually researching competitors, designing architecture, estimating costs, and planning sprints—you describe what you want to build, answer a few questions, and let AI do the heavy lifting.

The 4-step process:

1. 📝 Fill Template - Open a form and describe your project (like filling out a survey)
2. ✓ Validate Setup - System checks you have everything needed (takes 30 seconds)
3. ⚙️ Configure Options - Answer 8 simple questions about how you want the plan made
4. ✨ Generate Plan - AI creates your complete project plan (30 min - 2 hours depending on depth)

What you get: A folder with 50+ documents including market research, architecture diagrams, cost breakdowns, sprint plans, and implementation roadmaps—everything you'd normally spend weeks creating manually.

---

The 8 Questions (No Flags to Remember!)

Instead of memorizing command-line flags, you answer 8 interactive question groups that cover every feature: