# Multi-Agent Coding **A systematic methodology for building complex software with coordinated AI agents.** This repository demonstrates PRD-first development using Claude Code's multi-agent orchestration to build applications faster through parallel execution and contract-driven development. ## What This Is This project teaches a reusable process for AI-assisted development: 1. **Start with a comprehensive PRD** - Clear requirements enable everything else 2. **Configure CLAUDE.md** - Point to your PRD and planning agent 3. **Define specialized agents** - Each owns a specific domain with clear boundaries 4. **Execute in parallel** - Multiple agents work simultaneously 5. **Integrate smoothly** - Contracts ensure clean composition ## Repository Structure \`\`\` multi-agent-coding/ ├── agent-templates/ # Reusable templates for your projects │ ├── CLAUDE.md.example # Configuration template │ └── subagent-template.md # Domain agent template │ └── blackjack-app/ # Complete working example ├── blackjack-prd.md # Starting point: comprehensive PRD ├── CLAUDE.md # Agent orchestration config └── .claude/agents/ # Agent definitions ├── 01-foundation.md ├── 02-logic.md ├── 03-frontend.md └── 04-integration.md \`\`\` ## Quick Start ### Prerequisites - [Claude Code](https://docs.claude.com/en/docs/claude-code/overview) installed - Claude Pro subscription ### Try the Demo **Build the blackjack app with Claude:** \`\`\`bash # Clone the repository git clone https://github.com/bufothefrog/multi-agent-coding.git cd multi-agent-coding/blackjack-app cp .claude/settings.local.json.example .claude/settings.local.json claude # Tell Claude: "Please implement the blackjack-prd.md" \`\`\` Claude will read the PRD, discover the agent definitions in \`.claude/agents/\`, and orchestrate the build using specialized agents working in parallel. **Explore the methodology:** \`\`\`bash # 1. Read the PRD (the starting point) cat blackjack-app/blackjack-prd.md # 2. Review the configuration cat blackjack-app/CLAUDE.md # 3. Examine agent definitions cat blackjack-app/.claude/agents/01-foundation.md \`\`\` **Adapt to your project:** \`\`\`bash cp agent-templates/CLAUDE.md.example your-project/CLAUDE.md cp agent-templates/planning-agent.md your-project/.claude/agents/planning-agent.md # Customize using subagent-template.md \`\`\` ## Core Principles 1. **PRD First** - Comprehensive requirements enable parallel development 2. **Contracts Enable Parallelism** - Define interfaces before implementation 3. **Clear Boundaries** - Each agent owns a specific domain 4. **Context Management** - Chunk work to respect token limits 5. **Easy Integration** - Smooth composition validates the approach ## Why This Works **Traditional Sequential:** - Build backend → Build frontend → Add features → Debug integration - 3-4 weeks, tight coupling, painful integration **Multi-Agent Parallel:** - Define contracts → [Backend + Frontend + Features simultaneously] → Integration - 1-2 weeks, loose coupling, smooth integration ## Resources - [Advanced Context Engineering for Coding Agents](https://github.com/humanlayer/advanced-context-engineering-for-coding-agents/blob/main/ace-fca.md) - Comprehensive guide to context engineering techniques for AI coding agents