Claude Agents Repository

A production-ready, multi-runtime agent template system that generates artifacts for Claude Code Subagents, LangGraph, and OpenAI Assistants from a single YAML specification.

🚀 Quick Start

# 1. Clone and install
git clone <repository-url>
cd claude-code-agents
pip install -e ".[dev]"

# 2. Validate existing agents
agents validate

# 3. Generate Claude Code subagents
agents render claude --install

# 4. Generate LangGraph agents
agents render langgraph

# 5. List available agents
agents list-agents

📋 What's Inside

• agents/ — YAML specifications organized by domain (engineering, data, ops, product)
• recipes/ — Multi-agent workflows for complex tasks
• adapters/ — Runtime-specific generators (Claude, LangGraph, Assistants)
• templates/ — Jinja2 templates for artifact generation
• schemas/ — JSON Schema validation for specifications
• scripts/ — CLI tools for validation, rendering, and management

🏗️ Architecture

Single Source of Truth

Write once in YAML, deploy everywhere:

id: security-auditor
name: Security Auditor  
summary: Reviews code for vulnerabilities using OWASP standards
role: |
  You are a senior security engineer specializing in application security...

model:
  provider: anthropic
  family: claude
  tier: sonnet
  
tools:
  - id: code_scanner
    type: mcp
    spec: tools/mcp_servers/security_scanner.yaml

Multi-Runtime Support

Claude Code Subagent → ~/.claude/agents/security-auditor.md

---
name: security-auditor
description: Reviews code for vulnerabilities using OWASP standards
model: sonnet
tools: code_scanner
---

You are a senior security engineer...

LangGraph Agent → adapters/langgraph/security_auditor_agent.py

from langgraph.prebuilt import create_react_agent

class SecurityAuditorAgent:
    def __init__(self):
        self.model = ChatAnthropic(model="claude-3-5-sonnet-20241022")
        # ... implementation

🔧 CLI Commands

Validation

# Validate all specifications
agents validate

# Validate specific file
agents validate agents/engineering/security-auditor.yaml

# Verbose validation with detailed errors
agents validate --verbose

Rendering

# Generate Claude subagents 
agents render claude

# Generate and install to ~/.claude/agents
agents render claude --install

# Generate LangGraph agents
agents render langgraph

# Render specific agent
agents render claude --agent-id security-auditor

Management

# List all agents
agents list-agents

# Filter by domain
agents list-agents --domain engineering

# Filter by tag
agents list-agents --tag security

# Create new agent from template
agents init-agent my-new-agent --domain custom --name "My New Agent"

📖 Agent Specification

Agents follow the Agent Spec v1 schema:

Required Fields

• id: Unique identifier (kebab-case)
• name: Human-readable name
• summary: Brief capability description
• role: System prompt defining behavior
• model: Provider, family, and tier configuration
• ownership: Owner contact information
• version: Semantic version

Optional Fields

• invocation: Auto-trigger or explicit invocation
• tools: MCP servers, HTTP APIs, built-in functions
• constraints: Token limits, cost budgets, PII policies
• memory: Ephemeral, persistent, or hybrid storage
• evaluation: Acceptance criteria and test cases
• observability: Logging, tracing, and metrics

🔄 Multi-Agent Workflows

Recipes define complex workflows combining multiple agents:

id: feature-development
name: Complete Feature Development Workflow
graph:
  - stage: planning
    sequence:
      - agent: business-analyst
      - agent: backend-architect
        
  - stage: security_review
    sequence:
      - agent: security-auditor
        
  - stage: implementation
    parallel:
      - agent: docs-architect
      - agent: test-automator

🏢 Available Agents

Engineering

• backend-architect — Scalable system design and API architecture
• security-auditor — OWASP/CIS compliance and vulnerability assessment
• test-automator — Test strategy, automation, and coverage analysis

Data

• data-engineer — ETL pipelines, data quality, and warehouse design

Operations

• incident-responder — Structured incident response and post-mortems

Product

• business-analyst — Requirements gathering and user story creation
• docs-architect — Technical documentation and API guides

🛡️ Security & Quality

Built-in Safeguards

• Schema validation for all YAML specifications
• Secret detection in pre-commit hooks and CI
• PII policies configurable per agent
• Security scanning with Bandit and Safety

Quality Gates

• Automated testing with pytest (>80% coverage)
• Code formatting with Ruff and Black
• YAML linting with yamllint
• Type checking with mypy

🚀 Getting Started

1. Installation

git clone <repository-url>
cd claude-code-agents
pip install -e ".[dev]"

2. Set up development environment

# Install pre-commit hooks
pre-commit install

# Run tests
pytest

# Validate all specs
agents validate

3. Create your first agent

# Initialize from template
agents init-agent my-expert --domain engineering --name "My Expert"

# Edit the generated file
vim agents/engineering/my-expert.yaml

# Validate and render
agents validate agents/engineering/my-expert.yaml
agents render claude --agent-id my-expert --install

4. Deploy to your runtime

Claude Code: Agents are automatically installed to ~/.claude/agents/ with --install flag

LangGraph:

agents render langgraph
cd adapters/langgraph
pip install -r requirements.txt
python app.py  # Starts FastAPI server on :8000

📊 Success Metrics

• Time-to-first-agent: ≤ 10 minutes from clone to working subagent
• Time-to-first-workflow: ≤ 20 minutes for 3-stage recipe
• Test coverage: ≥ 80% for critical paths
• Zero secrets committed to repository

📚 Documentation

• Architecture Overview
• Agent Style Guide
• Security Policies
• Contribution Guidelines

🤝 Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/my-agent)
3. Follow the style guide for prompts and naming
4. Add tests for new functionality
5. Run the full validation suite (make test)
6. Submit a pull request

📄 License

Apache-2.0 — see LICENSE for details.

🆘 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Docs: Full Documentation

---

⚡ Built for production • 🔒 Security-first • 🚀 Developer-friendly