A multi-agent development methodology for AI-powered software development using specialized agents that coordinate through structured workflows to ensure security, quality, and velocity.
ClaudeCraft is a systematic approach to AI development that coordinates five specialized agents to deliver production-ready software. It's not a traditional software framework but rather a comprehensive methodology with:
- Structured agent coordination with mandatory handoff procedures
- Comprehensive tracking system for decisions, progress, and issues
- Quality gates ensuring security, performance, and maintainability
- Token management preventing context loss during development
- Natural language interface supporting conversational development
| Agent | Role | Expertise | Session Type |
|---|---|---|---|
| Planner | Strategic planning, PRD creation, architecture | Rich persistent memory | Long-running |
| Coder | Feature implementation, testing, technical setup | Lightweight task-focused | Task-focused |
| Tester | Quality assurance, bug discovery, performance | Feature-specific testing | Verification |
| Documenter | Documentation, user guides, API docs | Content creation | Documentation sprints |
| Reviewer | Code quality, security audits, architecture review | Comprehensive analysis | Review sessions |
- One feature at a time - Complete implementation before starting new work
- Token awareness - Stop at 75% context limit with structured handoffs
- Security first - OWASP Top 10 compliance from day one
- Mainstream technologies only - React/Next.js, Python/FastAPI, Node.js
-
Clone or use as template:
git clone https://github.com/your-org/claudecraft.git your-project cd your-project -
Review workspace structure:
workspace/ prd.md # Product Requirements Document todo.md # Sprint tasks and priorities memory-bank.md # Architectural decisions progress-log.md # Development tracking bug-tracker.md # Issue management agent-instructions/ # Agent-specific role definitions templates/ # Standard templates handoffs/ # Agent transition management -
Start with natural language:
"I have an idea for a task management app for small teams" "Fix the login bug users are reporting" "Add dark mode to the dashboard" "Review the security of our authentication"
- React + Python: Frontend with React/Next.js, backend with Python/FastAPI
- React + Node: Frontend with React/Next.js, backend with Node.js/Express
- React Only: Single-page applications with React/Next.js
Standard Feature Development:
Planner → Coder → Tester → Documenter → Reviewer
Bug Fix Cycle:
Tester → Coder → Tester → Reviewer
Architecture Planning:
Planner → Reviewer → Planner → Coder
Every agent transition requires:
- ✅ Update all workspace tracking files
- ✅ Create comprehensive handoff document
- ✅ Ensure tests passing and code stable
- ✅ Archive previous handoff
- ✅ Define clear next steps and success criteria
- 80%+ test coverage for all new code
- 300 lines maximum per file
- OWASP Top 10 compliance verified
- Sub-200ms API responses, sub-3s page loads
- Comprehensive documentation with every change
- Complete Workflow Guide - How the agent system actually works
- Workspace Guide - Understanding the workspace structure
- Agent Instructions - Detailed role definitions for each agent
- Best Practices - Proven patterns for development velocity and quality
- Security Guidelines - Security-first development approach
- >80% test coverage across all components
- OWASP Top 10 security compliance verified
- <200ms API response times and <3s page loads
- >95% successful agent handoffs with minimal clarification needed
MIT License - see LICENSE for details.