Claude Agent Framework is a battle-tested system that transforms Claude Code into a team of specialized AI agents working in parallel on your project. Whether you're building a startup MVP, enterprise system, or personal project, this framework adapts to your stack and your patterns.

One prompt. Two minutes. Full agent team.

# In your project directory
$ claude

# Paste this:
"I want to set up Claude Agent Framework for my project.
Please read the SYSTEM_GENERATOR_PROMPT.md from ./claude-agent-framework/
and create my custom agent system."

# That's it. You're done.

Your custom agent team is now ready with:

• Agents that understand your tech stack
• Commands tailored to your workflow
• Patterns extracted from your codebase
• Parallel execution for 3-6x speed boost

Run the generator. Get instant:

• /build - Parallel TDD development
• /debug - Smart debugging
• /test - Comprehensive testing
• /deploy - One-command deployment

Your system learns and adapts:

• Agents understand your code patterns
• Commands evolve with your workflow
• Context stays minimal (<10KB)
• Speed increases as patterns emerge

Build your dream team:

• Add specialized agents for new domains
• Create custom workflows for your process
• Integrate with your CI/CD pipeline
• Scale to any project size

• Manual Claude setup takes hours of context building
• Sequential execution wastes time on independent tasks
• Context overload (250KB+) slows every interaction
• Repetitive prompts for common workflows

• 2-minute setup with intelligent auto-configuration
• Parallel agents work simultaneously (3-6x faster)
• Minimal context (<10KB auto-loaded)
• Smart commands that orchestrate complex workflows

your-project/
├── .claude/                    # Ultra-light core (<10KB)
│   ├── agent-launcher.md      # Your mission control
│   ├── settings.json          # Project metadata
│   ├── MEMORY.md              # Cross-conversation memory
│   ├── agents/                # Custom subagent types
│   ├── rules/                 # Path-specific rules
│   └── commands/              # Your power tools
│       ├── build.md          # Feature development
│       ├── debug.md          # Problem solving
│       ├── test.md           # Quality assurance
│       └── deploy.md         # Ship to production
│
└── .claude-library/           # On-demand specialists
    ├── REGISTRY.json         # Central configuration (v2.0)
    ├── agents/               # Your AI team
    ├── contexts/             # Shared knowledge
    └── skills/               # Skill definitions

Claude Agent Framework automatically adapts to:

Frontend: React, Vue, Angular, Svelte, Next.js, Nuxt, SvelteKit Backend: Node.js, Python, Go, Rust, Java, C#, Ruby Databases: PostgreSQL, MySQL, MongoDB, Redis, SQLite Cloud: AWS, GCP, Azure, Vercel, Netlify, Cloudflare Testing: Jest, Pytest, Playwright, Cypress, Vitest AI/ML: OpenAI, Anthropic, LangChain, HuggingFace, LocalLLMs

"Start simple. Ship fast. Scale infinitely."

1. Progressive Enhancement - Start with basics, add complexity as needed
2. Context Minimalism - Load only what's essential
3. Parallel Everything - Why wait when agents can work together?
4. Your Patterns First - Adapt to your code, not the other way around

claude> Use Claude Agent Framework SYSTEM_GENERATOR_PROMPT.md to set up my project

claude> Follow AGENT_SYSTEM_TEMPLATE.md for manual configuration

claude> Study CLAUDE_AGENT_FRAMEWORK.md and build custom system

• 9 files archived, 50% fewer root docs
• REGISTRY.json slimmed 64% (1,154 -> 421 lines)
• 10+ new Claude Code features integrated

• Agent Teams: Custom subagent types in .claude/agents/
• Path-Specific Rules: .claude/rules/ with glob-based targeting
• Auto Memory: Persistent cross-conversation knowledge via MEMORY.md
• Extended Context: 1M token context window
• MCP Tool Search: Deferred tool loading via ToolSearch
• Effort Levels: model: "haiku" for fast/cheap, model: "opus" for complex
• Worktree Isolation: isolation: "worktree" for safe parallel work
• Background Agents: run_in_background: true for async execution
• Managed Settings: Centralized permission and model configuration
• Skills with context:fork: Branch context for skill execution

Track agent workflows with local SQLite-based monitoring. Zero cloud dependencies. Learn more ->

Add deterministic control: auto-format, block dangerous operations, validate outputs. Learn more ->

Both patterns disabled by default. Zero overhead when off.

Claude Agent Framework is built from real-world patterns used in production:

• Share your patterns - Add to AGENT_PATTERNS.md
• Submit improvements - Make it better for everyone
• Report issues - Help us fix what's broken

MIT License - Use freely in your projects, commercial or otherwise.

$ claude
> "Set up Claude Agent Framework for my project"