Skip to content

nategarelik/agentic-startup-claude-config

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

The Agentic Startup - Claude Code Configuration

Ship faster. Ship better. Ship with The Agentic Startup.

License: MIT Claude Code Version Sonnet 4.5 Activity-Based

A high-energy, spec-driven Claude Code configuration that transforms Claude into your AI technical co-founder. Built on activity-based agent collaboration, parallel execution, and enterprise-grade safety.

What is The Agentic Startup?

The Agentic Startup is a revolutionary development methodology that combines:

  • 11 Role Archetypes covering all aspects of software development
  • 5 Specialized Subagents for deep expertise
  • 39 Activities spanning the entire development lifecycle
  • Parallel Execution for maximum efficiency
  • GitHub Spec Kit integration for comprehensive specifications
  • Enterprise Safety with auto-approve + deny-list protection

Core Philosophy

  • Activity-Based, Not Role-Based: Agents selected by what they DO, not titles
  • Parallel by Default: Multiple agents working simultaneously
  • Spec-Driven Development: Comprehensive specs before code
  • Continuous Validation: Phase gates ensure quality
  • Progressive Enhancement: Iterative improvement cycles
  • Knowledge Capture: Document patterns and learnings

Features

High-Energy Development

  • Rapid iteration with continuous validation
  • Startup speed with production-ready quality
  • Multi-agent parallel execution
  • Intelligent task routing and orchestration

Comprehensive Agent System

  • 11 Role Archetypes: Covering every development activity
  • 5 Specialized Subagents: For complex, domain-specific tasks
  • Activity Router: Intelligent delegation based on task type
  • GitHub Copilot Integration: Full access to AI-powered development

Enterprise Safety

  • Auto-Approved Operations: Read files, git commits, PR creation
  • Deny-List Protection: 20+ destructive operations blocked
  • Permission Framework: Granular control over operations
  • Security Auditing: All operations logged for compliance

Token Resilience

  • Artifact Offload: Bulky content saved to disk
  • Checkpoint System: Recovery points at phase gates
  • Context Management: Automatic summarization
  • Resume Capability: Continue from any checkpoint

The 11 Role Archetypes

Chief - Orchestration & Leadership

Activities: Complexity assessment, work routing, bottleneck elimination, parallel planning Use: "Chief: Assess complexity and route this work"

Analyst - Requirements & Research

Activities: Requirements clarification/documentation, feature prioritization, solution research, project coordination Use: "Analyst: Clarify and document these requirements"

Architect - System Design

Activities: System design/documentation, architecture/code review, scalability planning, technology evaluation/standards Use: "Architect: Design the system architecture"

Software Engineer - Core Development

Activities: API design/documentation, database design, service integration, component architecture, business logic, reliability engineering, performance optimization, state management, browser compatibility Use: "Software Engineer: Implement the API and business logic"

Platform Engineer - Infrastructure

Activities: System performance, observability, containerization, pipeline engineering, CI/CD automation, deployment strategies, incident response, infrastructure as code, storage architecture, query optimization, data modeling Use: "Platform Engineer: Set up CI/CD and deployment"

Designer - User Experience

Activities: Accessibility implementation, user research, interaction design, visual design, design systems, information architecture Use: "Designer: Create the user interface design"

QA Engineer - Testing

Activities: Test strategy, test implementation, exploratory testing, performance testing Use: "QA Engineer: Develop comprehensive test strategy"

Security Engineer - Protection

Activities: Vulnerability assessment, authentication systems, security incident response, compliance audit, data protection Use: "Security Engineer: Audit and secure the system"

Mobile Engineer - Mobile Development

Activities: Mobile interface design, mobile data persistence, cross-platform integration, mobile deployment, mobile performance Use: "Mobile Engineer: Build the mobile experience"

ML Engineer - AI/ML Operations

Activities: Model deployment, ML monitoring, prompt optimization, MLOps automation, context management, feature engineering Use: "ML Engineer: Deploy and optimize ML models"

Meta Agent - Agent Evolution

Activities: Agent generation, agent validation, agent refactoring, evidence-based agent design Use: "Meta Agent: Optimize our agent collaboration"

The 5 Specialized Subagents

Opus Planner

Purpose: Complex planning with deep reasoning Model: Claude Opus 4.1 Activation: Complexity score > 8 Output: Comprehensive PLAN.md

Config Maintainer

Purpose: Manages .claude/** configuration Features: Context overflow handling, checkpoint system, agent evolution Activation: Meta Agent activities

Unity Builder

Purpose: Unity development specialist Features: Path restrictions, Unity Test Framework, CI/CD generation Activation: Unity-specific activities

Researcher

Purpose: File-first research approach Features: Citation management, domain analysis, pattern discovery Activation: Research activities

Life-Ops

Purpose: Email/calendar management via MCP Features: Gmail & Google Calendar OAuth, inbox triage, agenda generation Activation: Email/calendar operations

Quick Start

Installation

  1. Clone this repository to your Claude config directory:

    cd ~/.claude
    git clone https://github.com/nategarelik/agentic-startup-claude-config .
  2. Or copy the configuration files:

    cp -r agentic-startup-claude-config/.claude/* ~/.claude/
  3. Verify installation:

    ls ~/.claude/CLAUDE.md

Basic Usage

Once installed, Claude will automatically use The Agentic Startup configuration. Simply use the commands:

# Transform an idea into comprehensive specs
/s:specify "user authentication system"

# Execute phase-by-phase with approval gates
/s:implement

# Improve code quality without breaking
/s:refactor "payment processing module"

# Discover and document system knowledge
/s:analyze "authentication system"

Example Workflow

# 1. Specify a feature
User: "/s:specify task management system with real-time updates"

# Claude activates:
# - The Analyst (clarifies requirements)
# - The Architect (designs system)
# - Generates PRD, SDD, PLAN documents

# 2. Implement the feature
User: "/s:implement"

# Claude orchestrates parallel execution:
# - Software Engineer: Builds APIs
# - Platform Engineer: Sets up infrastructure
# - Designer: Creates UI components
# - QA Engineer: Implements tests
# - Security Engineer: Audits security

# 3. Refine and optimize
User: "/s:refactor"

# Multiple agents collaborate to improve code quality

Commands Reference

The Agentic Startup Commands

Command Description Example
/s:specify [idea] Generate PRD, SDD, and implementation plan /s:specify "real-time chat"
/s:implement Execute phase-by-phase with approval gates /s:implement
/s:refactor [target] Improve code quality without breaking /s:refactor "database layer"
/s:analyze [system] Discover and document system knowledge /s:analyze "auth flow"

GitHub Spec Kit Commands

Command Description Example
constitution Create/update project principles constitution
clarify Ask clarification questions clarify
plan [tech stack] Create technical implementation plan plan "Node.js + React"
tasks Generate task breakdown with parallel optimization tasks

Execution Commands

Command Description Example
parallel Execute multiple agents/tracks simultaneously parallel
status Report current progress across all agents status
checkpoint Save current state for recovery checkpoint "after-api"
review Multi-agent code quality and security analysis review
ship Prepare for production deployment ship

Agent Delegation

Directly invoke specific agents for targeted tasks:

# Role Archetypes
"Chief: Assess complexity and route this work"
"Analyst: Research authentication solutions"
"Architect: Design microservices architecture"
"Software Engineer: Implement payment API"
"Platform Engineer: Configure Kubernetes cluster"
"Designer: Create dashboard interface"
"QA Engineer: Generate test suite"
"Security Engineer: Audit API endpoints"
"Mobile Engineer: Build React Native app"
"ML Engineer: Deploy recommendation model"
"Meta Agent: Optimize agent collaboration"

# Specialized Subagents
"Opus Planner: Create complex implementation plan"
"Config Maintainer: Update agent configuration"
"Unity Builder: Create 3D game mechanics"
"Researcher: Analyze competitor solutions"
"Life-Ops: Triage my inbox"

Parallel Execution Strategy

The Agentic Startup organizes work into parallel tracks for maximum efficiency:

Track A: Backend Infrastructure

  • Database setup and migrations
  • Core business logic
  • API endpoints
  • Authentication/authorization

Track B: Frontend Development

  • UI components
  • State management
  • User interactions
  • Responsive design

Track C: Integration

  • API integration
  • External services
  • Data synchronization
  • Real-time features

Track D: Testing

  • Unit tests (80% coverage minimum)
  • Integration tests
  • E2E tests
  • Performance tests

Track E: DevOps & Documentation

  • CI/CD pipeline
  • Deployment configuration
  • API documentation
  • User guides

Track F: Version Control

  • Branch management
  • Commit orchestration
  • PR creation and management
  • Merge coordination

Model Policy

Default Model: Claude Sonnet 4.5

  • Use for: All agents, coding, implementation, refactoring, analysis
  • Why: Best coding model in the world, 17% improvement
  • Performance: Fast, accurate, comprehensive

Planning Model: Claude Opus 4.1

  • Use for: Complex planning when ambiguity is high (complexity score > 8)
  • Activation: Single-pass planning → outputs PLAN.md → handoff to Sonnet
  • Why: Deep reasoning for complex architectural decisions

Extended Thinking

  • Default: OFF
  • Enable: Per-task with small budget (5000ms) for auditable reasoning
  • Use cases: Security audits, architecture reviews, complex debugging

Safety & Permissions

Auto-Approved Operations

  • Read files, grep, glob
  • Git branch, commit, PR creation
  • Write to sandbox & project directories
  • Whitelisted commands: npm, python, docker, Unity, gh copilot, etc.

Requires User Approval

  • Non-whitelisted shell commands
  • Network operations
  • CI/CD pipeline changes
  • Writes outside allowed paths
  • File deletions

Hard Deny (Safety Boundaries)

  • git push --force
  • rm -rf /*
  • terraform destroy
  • kubectl delete namespace
  • Plus 15+ more destructive operations

Result: Feels permissionless day-to-day, but actually protected.

Context Management

Artifact Offload Strategy

  1. Bulky content (logs, code, quotes) → .claude/state/artifacts/
  2. Summarize back into context
  3. On context overflow: trim, summarize, resume
  4. Checkpoints enable recovery

Checkpoint System

# Save recovery point
/s:checkpoint "after-database-setup"

# Resume from checkpoint
/s:resume "after-database-setup"

Artifact Types

  • code-[feature]-[timestamp].md
  • logs-[component]-[timestamp].txt
  • data-[source]-[timestamp].json
  • quotes-[source]-[timestamp].md

GitHub Copilot Integration

The Agentic Startup has FULL access to GitHub Copilot CLI:

Code Operations

  • Generate any code, in any language, for any framework
  • Explain complex code, errors, and patterns
  • Suggest fixes and optimizations
  • Refactor for quality improvements

GitHub Operations

  • Create feature branches with meaningful names
  • Generate intelligent commit messages
  • Auto-generate PR descriptions with full context
  • Review PRs with AI-powered feedback

DevOps Operations

  • Generate CI/CD workflows
  • Create Dockerfiles and Kubernetes configs
  • Suggest infrastructure as code
  • Optimize deployment strategies

Intelligence Operations

  • Analyze performance bottlenecks
  • Security audit code
  • Suggest architecture improvements
  • Generate migration plans

File Structure

project/
├── .claude/
│   ├── CLAUDE.md                    # Main configuration
│   ├── agents/                      # Specialized agent definitions
│   │   ├── opus-planner.md
│   │   ├── config-maintainer.md
│   │   ├── unity-builder.md
│   │   ├── researcher.md
│   │   └── life-ops.md
│   ├── policies/                    # Permission and security
│   │   ├── permissions.json
│   │   └── security.json
│   ├── commands/                    # Custom command definitions
│   └── state/                       # Runtime state (gitignored)
│       ├── artifacts/
│       ├── thinking/
│       ├── summaries/
│       └── checkpoints/
├── specs/                           # Feature specifications
│   └── [feature-name]/
│       ├── PRD.md
│       ├── SDD.md
│       ├── PLAN.md
│       ├── tasks.md
│       └── contracts/
├── docs/                            # Documentation
│   ├── domain/                      # Business domain
│   ├── patterns/                    # Discovered patterns
│   └── interfaces/                  # API and interface docs
└── src/                             # Implementation files

Success Metrics

The Agentic Startup aims to achieve:

  • 80%+ test coverage
  • < 200ms API response times
  • Zero critical security issues
  • 90%+ documentation completeness
  • < 1% production error rate
  • 10x faster specification-to-implementation
  • 5x reduction in context switches

Quality Standards

Code Quality

  • Clean, readable code with meaningful names
  • Comprehensive error handling
  • Performance-optimized implementations
  • Security best practices
  • Accessibility compliance

Testing

  • Unit tests for all business logic
  • Integration tests for APIs
  • E2E tests for critical user paths
  • Performance tests for bottlenecks
  • Security tests for vulnerabilities

Documentation

  • Clear README with setup instructions
  • API documentation with examples
  • Inline comments for complex logic
  • Architecture decision records
  • Deployment guides

Examples

Example 1: Building a Feature

User: "Build user authentication with JWT"

# Phase 1: Specification
Claude: "Analyst: Clarifying requirements..."
# Generates PRD, SDD, PLAN

# Phase 2: Implementation (Parallel)
Claude: "Orchestrating parallel execution..."
# - Software Engineer: Builds API
# - Designer: Creates login UI
# - QA Engineer: Writes tests
# - Security Engineer: Audits security
# - Platform Engineer: Sets up CI/CD

# Phase 3: Delivery
Claude: "All tracks complete. Creating PR..."
# Auto-generates comprehensive PR description

Example 2: Refactoring

User: "/s:refactor database layer"

Claude: "Architect: Analyzing current implementation..."
# - Detects code smells
# - Suggests improvements
# - Applies changes incrementally
# - Runs tests to verify behavior
# - Documents changes

Example 3: Research & Analysis

User: "/s:analyze authentication flow"

Claude: "Researcher: Discovering patterns..."
# - Maps authentication flow
# - Documents business rules
# - Identifies security considerations
# - Creates knowledge repository
# - Generates documentation

MCP Integration (Life Operations)

Enabled Services

  • Gmail (@modelcontextprotocol/server-gmail)
  • Google Calendar (@modelcontextprotocol/server-google-calendar)

Security

  • OAuth2 authentication required
  • All operations logged
  • Rate limited (60/min Gmail, 30/min Calendar)
  • Minimal scopes
  • Never permanently deletes emails

Usage

"Life-Ops: Triage my inbox"
"Life-Ops: Show today's agenda"
"Life-Ops: Extract tasks from emails"

Contributing

We welcome contributions! Please see our Contributing Guide for details.

Ways to Contribute

  • Add new agent archetypes
  • Enhance existing agents
  • Improve documentation
  • Share example workflows
  • Report issues and bugs
  • Suggest new features

License

MIT License - see LICENSE for details.

Acknowledgments

Support


Version: 2.0.0 Status: Production Ready Model: Claude Sonnet 4.5 (default), Opus 4.1 (complex planning) Motto: "Ship faster. Ship better. Ship with The Agentic Startup."

Built with by the Claude Code community

About

The Agentic Startup - High-energy, spec-driven Claude Code configuration with 11 role archetypes, 5 specialized subagents, parallel execution, and enterprise safety

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors