Multi-Agent Development System

A Claude Code plugin providing 76 specialized AI agents with hierarchical orchestration, T1/T2 cost optimization, runtime verification, workflow compliance validation, and comprehensive support for 7 programming languages across 10+ technology stacks. Built for full-stack, multi-language development with progress tracking, parallel development capabilities, automated runtime testing, and PR-based workflows.

76 Specialized Agents

Planning Agents (3)

PRD Generator - Interactive PRD creation with technology stack selection
Task Graph Analyzer - Breaks PRD into tasks with dependency analysis
Sprint Planner - Organizes tasks into executable sprints

Orchestration Agents (5)

Sprint Orchestrator - Manages entire sprint execution with runtime verification (Sonnet)
Task Orchestrator - Coordinates single task with T1/T2 escalation (Sonnet)
Requirements Validator - Quality gate ensuring 100% criteria satisfaction and runtime checks (Sonnet)
Track Merger - Intelligently merges parallel development tracks using git worktrees (Sonnet)
Workflow Compliance - Validates orchestrators followed required workflows and generated all mandatory artifacts (Sonnet)

Database Agents (15) - Expanded 3x for multi-language support

Database Designer - Schema design and normalization (Sonnet)
Database Developer Python T1/T2 - SQLAlchemy + Alembic (Haiku/Sonnet)
Database Developer TypeScript T1/T2 - Prisma/TypeORM (Haiku/Sonnet)
Database Developer Java T1/T2 - JPA/Hibernate + Flyway (Haiku/Sonnet)
Database Developer C# T1/T2 - Entity Framework Core + migrations (Haiku/Sonnet)
Database Developer Go T1/T2 - GORM + migrate (Haiku/Sonnet)
Database Developer Ruby T1/T2 - ActiveRecord + migrations (Haiku/Sonnet)
Database Developer PHP T1/T2 - Eloquent/Doctrine + migrations (Haiku/Sonnet)

Backend Agents (22) - Expanded 3x for enterprise stacks

API Designer - RESTful API specifications with OpenAPI (Sonnet)
API Developer Python T1/T2 - FastAPI/Django/Flask endpoints (Haiku/Sonnet)
API Developer TypeScript T1/T2 - Express/NestJS/Fastify endpoints (Haiku/Sonnet)
API Developer Java T1/T2 - Spring Boot/Micronaut endpoints (Haiku/Sonnet)
API Developer C# T1/T2 - ASP.NET Core/Minimal API endpoints (Haiku/Sonnet)
API Developer Go T1/T2 - Gin/Echo/Fiber endpoints (Haiku/Sonnet)
API Developer Ruby T1/T2 - Rails/Sinatra/Grape endpoints (Haiku/Sonnet)
API Developer PHP T1/T2 - Laravel/Symfony/Slim endpoints (Haiku/Sonnet)
Backend Code Reviewers - Python, TypeScript, Java, C#, Go, Ruby, PHP variants (Sonnet)

Frontend Agents (4)

Frontend Designer - UI/UX component specifications (Sonnet)
Frontend Developer T1/T2 - React/Vue/Next.js/Angular implementation (Haiku/Sonnet)
Frontend Code Reviewer - Accessibility, performance, best practices (Sonnet)

Python Agents (2)

Python Developer Generic T1/T2 - CLI tools, scripts, utilities (Haiku/Sonnet)

Quality Agents (11) - Comprehensive quality assurance and runtime verification

Test Writer - Unit, integration, and e2e tests (Sonnet)
Security Auditor - OWASP Top 10 compliance (Sonnet)
Documentation Coordinator - Technical documentation (Sonnet)
Runtime Verifier - Application launch verification and manual testing documentation (Sonnet)
Performance Auditor Python - Python-specific performance optimization (Sonnet)
Performance Auditor TypeScript - TypeScript/Node.js performance optimization (Sonnet)
Performance Auditor Java - Java/Spring Boot performance optimization (Sonnet)
Performance Auditor C# - .NET performance optimization (Sonnet)
Performance Auditor Go - Go performance optimization (Sonnet)
Performance Auditor Ruby - Ruby/Rails performance optimization (Sonnet)
Performance Auditor PHP - PHP/Laravel performance optimization (Sonnet)

Scripting Agents (4) - NEW

PowerShell Developer T1/T2 - Windows automation, Azure, DSC (Haiku/Sonnet)
Shell Developer T1/T2 - Bash/Zsh automation scripts (Haiku/Sonnet)

DevOps Agents (4) - NEW

Docker Specialist - Containerization, multi-stage builds, optimization (Sonnet)
Kubernetes Specialist - K8s manifests, Helm, operators (Sonnet)
CI/CD Specialist - GitHub Actions, GitLab CI, Jenkins (Sonnet)
Terraform Specialist - Infrastructure as Code for AWS/Azure/GCP (Sonnet)

Infrastructure Agents (2) - NEW

Configuration Manager T1/T2 - Config files, secrets, env management (Haiku/Sonnet)

Mobile Agents (4) - NEW

iOS Developer T1/T2 - SwiftUI, UIKit, native iOS development (Haiku/Sonnet)
Android Developer T1/T2 - Kotlin, Jetpack Compose, native Android (Haiku/Sonnet)

Cost Optimization (T1/T2 System)

T1 Tier (Haiku) - Cost-optimized first attempt

Handles 70-80% of implementation work
$0.001 per 1K tokens
Automatic escalation on failure

T2 Tier (Sonnet) - Enhanced quality

Handles complex scenarios after T1 validation fails
15-20% of work requiring deeper analysis
$0.003 per 1K tokens

Result: Significant cost optimization through intelligent model selection

Quality Gates

100% Criteria Satisfaction - Requirements validator enforces acceptance criteria
Runtime Verification - Applications must launch without errors before completion
Automated Testing - All tests must pass (100% pass rate required, external API tests may be skipped)
Testing Documentation - TESTING_SUMMARY.md generated with complete test results
Workflow Compliance - Validates all required artifacts generated and processes followed
Security First - OWASP Top 10 audits on all implementations
80%+ Test Coverage - Enforced by test writer agent
Code Review - Language-specific reviewers for every implementation
Manual Testing Documentation - Comprehensive testing guides for human verification
Iterative Refinement - Max 5 iterations with T1→T2 escalation

Supported Languages & Frameworks

This system now provides development coverage across 7 major programming languages and 10+ development stacks:

Language	Frameworks	ORM/Database	Testing	Code Review
Python	FastAPI, Django, Flask	SQLAlchemy, Alembic	pytest, unittest	✅ Sonnet
TypeScript	Express, NestJS, Fastify	Prisma, TypeORM	Jest, Vitest	✅ Sonnet
Java	Spring Boot, Micronaut	JPA/Hibernate, Flyway	JUnit, TestNG	✅ Sonnet
C#	ASP.NET Core, Minimal API	Entity Framework Core	xUnit, NUnit	✅ Sonnet
Go	Gin, Echo, Fiber	GORM, migrate	testing, testify	✅ Sonnet
Ruby	Rails, Sinatra, Grape	ActiveRecord	RSpec, minitest	✅ Sonnet
PHP	Laravel, Symfony, Slim	Eloquent, Doctrine	PHPUnit, Pest	✅ Sonnet

Frontend:

React, Vue.js, Next.js, Angular (TypeScript)
Tailwind CSS, Material-UI, shadcn/ui
Vite, webpack, turbopack

Mobile:

iOS: SwiftUI, UIKit, Combine
Android: Kotlin, Jetpack Compose, Coroutines

Database:

PostgreSQL (primary), MySQL, SQLite, MongoDB

DevOps & Infrastructure:

Docker, docker-compose, Kubernetes
GitHub Actions, GitLab CI, Jenkins
Terraform (AWS, Azure, GCP)
PowerShell, Bash/Zsh scripting
Configuration management (env, YAML, JSON, secrets)

Development Stacks Supported

Full-Stack Web Applications:

MEAN/MERN Stack - MongoDB/Express/Angular or React/Node.js
Python Stack - FastAPI/Django + PostgreSQL + React
Java Enterprise - Spring Boot + PostgreSQL + Angular
C# .NET Stack - ASP.NET Core + SQL Server + React
Go Stack - Gin/Echo + PostgreSQL + Vue
Ruby Stack - Rails + PostgreSQL + React
PHP Stack - Laravel + MySQL + Vue

Mobile Applications: 8. iOS Native - SwiftUI + CoreData + REST APIs 9. Android Native - Kotlin + Room + Retrofit

DevOps & Infrastructure: 10. Cloud Infrastructure - Terraform + Docker + CI/CD 11. Microservices - Any language + Docker + Kubernetes config

Scripting & Automation: 12. System Administration - Bash/Python scripts + Terraform 13. Data Processing - Python + pandas + NumPy

New Capabilities

DevOps Integration

Containerization with Docker specialist (multi-stage builds, optimization)
Kubernetes orchestration specialist (Helm charts, operators, manifests)
CI/CD pipeline specialist for GitHub Actions, GitLab CI, Jenkins
Infrastructure as Code with Terraform specialist for AWS, Azure, GCP
Automated deployment workflows

Mobile Development

Native iOS development with SwiftUI and UIKit
Native Android development with Kotlin and Jetpack Compose
RESTful API integration and local data persistence
Platform-specific UI/UX best practices

Scripting & Automation

PowerShell for Windows automation and Azure management
Shell scripting for Linux/Unix system automation (Bash/Zsh)
Configuration management for env files, YAML, JSON, secrets
Build automation and deployment scripts
Database migration and backup scripts

Enterprise Language Support

Java enterprise applications with Spring Boot ecosystem
C# .NET Core applications for Windows/Linux/macOS
Go microservices with high-performance frameworks
Ruby web applications with Rails conventions
PHP applications with Laravel/Symfony frameworks

Installation

From GitHub (Recommended)

/plugin marketplace add https://github.com/michael-harris/claude-code-multi-agent-dev-system
/plugin install multi-agent-dev-system

From Local Path (Development)

# Clone the repository
git clone https://github.com/michael-harris/claude-code-multi-agent-dev-system.git
cd claude-code-multi-agent-dev-system

# Install locally
./install-local.sh

# OR manually
/plugin marketplace add file:///absolute/path/to/claude-code-multi-agent-dev-system
/plugin install multi-agent-dev-system

Verify Installation

/plugin list
# Should show: multi-agent-dev-system

Quick Start

Full Workflow

# 1. Generate PRD
/multi-agent:prd
# Interactive interview to create comprehensive PRD
# Output: docs/planning/PROJECT_PRD.yaml

# 2. Create tasks and sprints
/multi-agent:planning
# Breaks PRD into tasks with dependencies
# Organizes tasks into sprints
# Output: docs/planning/tasks/TASK-*.yaml
#         docs/sprints/SPRINT-*.yaml

# 3. Execute sprint
/multi-agent:sprint SPRINT-001
# Automated execution with quality loops
# T1 attempts first, escalates to T2 if needed
# Runtime verification ensures app launches without errors
# All automated tests must pass (100% pass rate)
# Manual testing guide generated for human verification
# Requirements validator ensures 100% criteria met

Individual Agent Usage

// Design database schema
Task(
  subagent_type="multi-agent:database:designer",
  model="opus",
  prompt="Design schema for user authentication with roles and permissions"
)

// Implement schema (T1 attempt)
Task(
  subagent_type="multi-agent:database:developer-python-t1",
  model="haiku",
  prompt="Implement the schema design from docs/api/database-schema.yaml using SQLAlchemy"
)

// If T1 fails validation, escalate to T2
Task(
  subagent_type="multi-agent:database:developer-python-t2",
  model="sonnet",
  prompt="Fix the implementation issues identified by the validator"
)

// Design API
Task(
  subagent_type="multi-agent:backend:api-designer",
  model="opus",
  prompt="Design REST API for user management"
)

// Security audit
Task(
  subagent_type="multi-agent:quality:security-auditor",
  model="opus",
  prompt="Audit the authentication implementation for OWASP Top 10 vulnerabilities"
)

Progress Tracking & Parallel Development

Progress Tracking (Resume Functionality)

All planning and execution workflows now include automatic progress tracking via state files:

State Files:

Projects: docs/planning/.project-state.yaml
Features: docs/planning/.feature-{id}-state.yaml
Issues: docs/planning/.issue-{id}-state.yaml

Resume Capabilities:

System tracks completion status of all tasks and sprints
Automatically skips completed work when resuming
Resume from any interruption point (system crash, manual stop, etc.)
State files provide detailed audit trail of development progress

Example:

# Start sprint execution
/multi-agent:sprint all

# ... system interrupted at SPRINT-003 ...

# Resume from where you left off
/multi-agent:sprint all
# System: "Resuming from SPRINT-003 (SPRINT-001, SPRINT-002 already complete)"

Parallel Development Tracks

Enable parallel development across independent task chains to dramatically reduce development time:

Single Track (Default):

/multi-agent:planning           # Creates sprints: SPRINT-001, SPRINT-002, SPRINT-003
/multi-agent:sprint all         # Sequential execution: ~128 hours

Parallel Tracks - State-Only Mode:

/multi-agent:planning 3         # Creates parallel tracks: SPRINT-001-01, SPRINT-001-02, SPRINT-001-03
                                # System calculates max possible tracks from dependencies
                                # Logical separation via state files

# Execute all tracks in parallel (different terminals/sessions):
/multi-agent:sprint all 01      # Terminal 1: Track 1 (~42 hours)
/multi-agent:sprint all 02      # Terminal 2: Track 2 (~48 hours)
/multi-agent:sprint all 03      # Terminal 3: Track 3 (~38 hours)

# Result: ~48 hours (62% faster than sequential)

Parallel Tracks - Git Worktrees Mode (NEW):

/multi-agent:planning 3 --use-worktrees
# Creates isolated git worktrees for true physical separation:
#  - .multi-agent/track-01/ (branch: dev-track-01)
#  - .multi-agent/track-02/ (branch: dev-track-02)
#  - .multi-agent/track-03/ (branch: dev-track-03)

# Each track works in its own directory - zero conflicts!
/multi-agent:sprint all 01      # Auto-switches to .multi-agent/track-01/
/multi-agent:sprint all 02      # Auto-switches to .multi-agent/track-02/
/multi-agent:sprint all 03      # Auto-switches to .multi-agent/track-03/

# After all tracks complete, merge them back:
/multi-agent:merge-tracks       # Intelligently merges all tracks, cleans up worktrees

# Expert commands (optional):
/multi-agent:worktree status    # Show detailed status of all worktrees
/multi-agent:worktree list      # List all worktrees with paths
/multi-agent:worktree cleanup   # Manually clean up worktrees

How It Works:

Task Graph Analyzer calculates max parallel tracks from dependency analysis
Sprint Planner distributes tasks across tracks using balanced bin-packing algorithm
Worktree Mode (optional): Creates isolated git worktrees + branches for each track
Track Merger (worktrees only): Intelligently merges tracks back to main branch
Each track contains independent task chains that can execute simultaneously
If requested tracks > max possible, system uses max and warns user

Benefits:

50-70% reduction in wall-clock development time
State-Only Mode: Simple, works in single directory, state files coordinate tracks
Worktree Mode: True isolation, no file conflicts, git-native separation
Ideal for projects with independent components (backend, frontend, infrastructure)
State tracking enables resumption for any track independently
Supports team collaboration (different tracks = different team members)

Worktree Mode Advantages:

Zero file conflicts - each track in separate directory
Better git history - separate branches with clear merge points
Safer parallel execution - no risk of state file conflicts
Natural merge workflow - git handles combining the work
Supports collaboration - team members can work in isolated worktrees

Worktree Management Commands

The system provides dedicated commands for managing git worktrees used in parallel development:

View detailed status:

/multi-agent:worktree status

Shows comprehensive status including:

Active worktrees with paths and branches
Number of commits in each track
Modified/staged files per worktree
Merge conflicts (if any)
Sprint progress per track
Recommendations for next steps

List all worktrees:

/multi-agent:worktree list

Simple tabular view of all worktrees with:

Worktree paths
Associated branches
Current HEAD commit
Lock status

Clean up worktrees:

/multi-agent:worktree cleanup [--delete-branches]

Safely removes worktrees:

Checks for uncommitted changes (warns and aborts if found)
Removes worktree directories
Optionally deletes associated branches with --delete-branches
Updates state files
Preserves main repository

Merge parallel tracks:

/multi-agent:merge-tracks [--keep-worktrees] [--delete-branches]

Intelligently merges all parallel development tracks:

Sequential merge strategy (track-01 → track-02 → track-03)
Intelligent conflict resolution for common patterns
Post-merge testing after each track integration
Automatic cleanup (unless --keep-worktrees specified)
Optional branch deletion with --delete-branches
Generates comprehensive merge completion report

Development History Preservation

Sprint and task files are intentionally preserved:

Provides complete development history and audit trail
Shows decision-making process and evolution of requirements
Valuable for debugging, rollback, and understanding context
Can be manually removed if desired, but recommended to keep
.gitignore can exclude state files (.*.yaml) if desired while keeping sprint definitions

Architecture

Hierarchical Orchestration

User → /multi-agent:prd, /multi-agent:planning, /multi-agent:sprint commands
   ↓
Sprint Orchestrator (Sonnet) - Manages entire sprint
   ↓
Task Orchestrator (Sonnet) - Coordinates single task with T1/T2 switching
   ↓
Specialized Agents (T1/T2) - Implement with automatic escalation
   ↓
Requirements Validator (Sonnet) - Quality gate (100% criteria met)
   ↓
Workflow Compliance (Sonnet) - Validates process adherence

T1→T2 Escalation Flow

Iteration 1: T1 agent (Haiku) attempts implementation
Iteration 2: T1 attempts fixes based on validator feedback
  → If PASS: Task complete ✅
  → If FAIL: Switch to T2 for iteration 3+

Iteration 3+: T2 agent (Sonnet) handles complexity

Max Iterations: 5 before human intervention required

Quality Gate System

Every sprint passes through:

Code Reviewers - Language-specific quality checks
Security Auditor - OWASP Top 10 compliance
Performance Auditors - Language-specific optimization
Runtime Verifier - Application launch verification and testing
- Builds and launches application (Docker or local)
- Runs all automated tests (100% pass rate required)
- Checks for runtime errors and exceptions
- Generates TESTING_SUMMARY.md with complete results
- Documents manual testing procedures for humans
Requirements Validator - Binary pass/fail on acceptance criteria
Workflow Compliance - Validates all required artifacts and processes
Test Writer - 80%+ coverage requirement

No sprint completes without 100% criteria satisfaction, successful runtime verification, and workflow compliance

Model Distribution

Model	Count	Use Cases	Cost/1K Tokens
Sonnet	54	T2 developers, reviewers, orchestration, code review, performance auditing, runtime verification, workflow compliance	$0.003
Haiku	22	T1 developers (first attempt across all languages)	$0.001

Total: 76 Agents

Note: All agents now use Sonnet or Haiku for cost optimization. Previous Opus-only agents (orchestrators, designers, security auditor) have been migrated to Sonnet with maintained quality standards.

Orchestration Intelligence:

5 orchestration agents (all Sonnet) coordinate the entire workflow
Sprint Orchestrator manages overall execution with 9-step process
Task Orchestrator handles individual tasks with T1→T2 escalation
Requirements Validator enforces 100% acceptance criteria satisfaction
Workflow Compliance ensures all mandatory artifacts generated
Track Merger handles parallel development track integration

Cost Optimization Logic:

T1 (Haiku) attempts implementation first (70-80% success rate)
T2 (Sonnet) handles complex cases after T1 failure (15-20% of work)
Sonnet for design decisions, orchestration, and quality gates

Multi-Language Scaling:

Each language gets T1/T2 developer pairs for database and backend work
Language-specific code reviewers ensure quality standards
Shared design agents (Sonnet) work across all languages
Same cost optimization applies to Java, C#, Go, Ruby, PHP as Python/TypeScript

Examples

Example 1: Complete Workflow

See examples/complete-workflow-example.md for a detailed walkthrough of building a task management application from PRD to deployment.

Example 2: Individual Agent Usage

See examples/individual-agent-usage.md for 10 targeted scenarios:

Quick database schema design
Implement database models from design
API design review
Implement API endpoint
Security audit
Generate tests for existing code
Create frontend component
Code review
Create CLI tool
Generate documentation

Example 3: Multi-Language Projects

See examples/multi-language-examples.md for examples demonstrating multi-language support:

Java Spring Boot - Enterprise REST API with JPA
C# .NET Core - Minimal API with Entity Framework
Go Microservice - High-performance API with GORM
Ruby on Rails - Convention-based web application
PHP Laravel - Full-stack web application
iOS Swift - Native mobile app with SwiftUI
Android Kotlin - Native mobile app with Jetpack Compose
DevOps Pipeline - Complete CI/CD with Docker and Terraform

Example 4: Parallel Development Tracks

See examples/parallel-tracks-example.md for a comprehensive walkthrough of using parallel development tracks to build an e-commerce platform with 50-70% reduction in wall-clock time.

Agent Reference

Complete list of all 76 agents organized by category:

Planning (3 agents)

planning:prd-generator (Sonnet)
planning:task-graph-analyzer (Sonnet)
planning:sprint-planner (Sonnet)

Orchestration (5 agents)

orchestration:sprint-orchestrator (Sonnet)
orchestration:task-orchestrator (Sonnet)
orchestration:requirements-validator (Sonnet)
orchestration:track-merger (Sonnet)
orchestration:workflow-compliance (Sonnet)

Database (15 agents)

database:designer (Sonnet)
database:developer-python-t1 (Haiku)
database:developer-python-t2 (Sonnet)
database:developer-typescript-t1 (Haiku)
database:developer-typescript-t2 (Sonnet)
database:developer-java-t1 (Haiku)
database:developer-java-t2 (Sonnet)
database:developer-csharp-t1 (Haiku)
database:developer-csharp-t2 (Sonnet)
database:developer-go-t1 (Haiku)
database:developer-go-t2 (Sonnet)
database:developer-ruby-t1 (Haiku)
database:developer-ruby-t2 (Sonnet)
database:developer-php-t1 (Haiku)
database:developer-php-t2 (Sonnet)

Backend (22 agents)

backend:api-designer (Sonnet)
backend:api-developer-python-t1 (Haiku)
backend:api-developer-python-t2 (Sonnet)
backend:api-developer-typescript-t1 (Haiku)
backend:api-developer-typescript-t2 (Sonnet)
backend:api-developer-java-t1 (Haiku)
backend:api-developer-java-t2 (Sonnet)
backend:api-developer-csharp-t1 (Haiku)
backend:api-developer-csharp-t2 (Sonnet)
backend:api-developer-go-t1 (Haiku)
backend:api-developer-go-t2 (Sonnet)
backend:api-developer-ruby-t1 (Haiku)
backend:api-developer-ruby-t2 (Sonnet)
backend:api-developer-php-t1 (Haiku)
backend:api-developer-php-t2 (Sonnet)
backend:code-reviewer-python (Sonnet)
backend:code-reviewer-typescript (Sonnet)
backend:code-reviewer-java (Sonnet)
backend:code-reviewer-csharp (Sonnet)
backend:code-reviewer-go (Sonnet)
backend:code-reviewer-ruby (Sonnet)
backend:code-reviewer-php (Sonnet)

Frontend (4 agents)

frontend:designer (Sonnet)
frontend:developer-t1 (Haiku)
frontend:developer-t2 (Sonnet)
frontend:code-reviewer (Sonnet)

Python (2 agents)

python:developer-generic-t1 (Haiku)
python:developer-generic-t2 (Sonnet)

Quality (11 agents)

quality:test-writer (Sonnet)
quality:security-auditor (Sonnet)
quality:documentation-coordinator (Sonnet)
quality:runtime-verifier (Sonnet)
quality:performance-auditor-python (Sonnet)
quality:performance-auditor-typescript (Sonnet)
quality:performance-auditor-java (Sonnet)
quality:performance-auditor-csharp (Sonnet)
quality:performance-auditor-go (Sonnet)
quality:performance-auditor-ruby (Sonnet)
quality:performance-auditor-php (Sonnet)

Scripting (4 agents) - NEW

scripting:powershell-developer-t1 (Haiku)
scripting:powershell-developer-t2 (Sonnet)
scripting:shell-developer-t1 (Haiku)
scripting:shell-developer-t2 (Sonnet)

DevOps (4 agents) - NEW

devops:docker-specialist (Sonnet)
devops:kubernetes-specialist (Sonnet)
devops:cicd-specialist (Sonnet)
devops:terraform-specialist (Sonnet)

Infrastructure (2 agents) - NEW

infrastructure:configuration-manager-t1 (Haiku)
infrastructure:configuration-manager-t2 (Sonnet)

Mobile (4 agents) - NEW

mobile:ios-developer-t1 (Haiku)
mobile:ios-developer-t2 (Sonnet)
mobile:android-developer-t1 (Haiku)
mobile:android-developer-t2 (Sonnet)

Repository Structure

claude-code-multi-agent-dev-system/
├── plugin.json              # Plugin manifest (76 agents, 10 commands)
├── README.md                # This file
├── LICENSE                  # MIT License
├── install-local.sh         # Local installation script
├── agents/                  # 76 agent definitions
│   ├── planning/           # PRD, task analysis, sprint planning (3)
│   ├── orchestration/      # Sprint/task orchestration, validation, workflow compliance, track merging (5)
│   ├── database/           # Schema design + 7-language implementation (15)
│   │   ├── designer.md
│   │   ├── developer-python-t1.md
│   │   ├── developer-python-t2.md
│   │   ├── developer-typescript-t1.md
│   │   ├── developer-typescript-t2.md
│   │   ├── developer-java-t1.md
│   │   ├── developer-java-t2.md
│   │   ├── developer-csharp-t1.md
│   │   ├── developer-csharp-t2.md
│   │   ├── developer-go-t1.md
│   │   ├── developer-go-t2.md
│   │   ├── developer-ruby-t1.md
│   │   ├── developer-ruby-t2.md
│   │   ├── developer-php-t1.md
│   │   └── developer-php-t2.md
│   ├── backend/            # API design + 7-language implementation + review (22)
│   │   ├── api-designer.md
│   │   ├── api-developer-[language]-t1.md (7 languages)
│   │   ├── api-developer-[language]-t2.md (7 languages)
│   │   └── code-reviewer-[language].md (7 languages)
│   ├── frontend/           # UI design + implementation + review (4)
│   ├── python/             # Generic Python development (2)
│   ├── quality/            # Testing, security, documentation (3)
│   ├── scripting/          # PowerShell and Shell scripting (4) - NEW
│   │   ├── powershell-developer-t1.md
│   │   ├── powershell-developer-t2.md
│   │   ├── shell-developer-t1.md
│   │   └── shell-developer-t2.md
│   ├── devops/             # Docker, Kubernetes, CI/CD, Terraform (4) - NEW
│   │   ├── docker-specialist.md
│   │   ├── kubernetes-specialist.md
│   │   ├── cicd-specialist.md
│   │   └── terraform-specialist.md
│   ├── infrastructure/     # Configuration management (2) - NEW
│   │   ├── configuration-manager-t1.md
│   │   └── configuration-manager-t2.md
│   ├── quality/            # Testing, security, documentation, performance, runtime verification (11)
│   │   ├── test-writer.md
│   │   ├── security-auditor.md
│   │   ├── documentation-coordinator.md
│   │   ├── runtime-verifier.md
│   │   ├── performance-auditor-python.md
│   │   ├── performance-auditor-typescript.md
│   │   ├── performance-auditor-java.md
│   │   ├── performance-auditor-csharp.md
│   │   ├── performance-auditor-go.md
│   │   ├── performance-auditor-ruby.md
│   │   └── performance-auditor-php.md
│   └── mobile/             # iOS and Android (4) - NEW
│       ├── ios-developer-t1.md
│       ├── ios-developer-t2.md
│       ├── android-developer-t1.md
│       └── android-developer-t2.md
├── commands/               # 10 workflow commands
│   ├── prd.md             # Generate PRD
│   ├── planning.md        # Create tasks and sprints (with parallel tracks support)
│   ├── sprint.md          # Execute single sprint
│   ├── sprint-all.md      # Execute all sprints (with track filtering)
│   ├── feature.md         # Complete feature workflow (PRD → Planning → Implementation)
│   ├── issue.md           # Complete issue resolution workflow
│   ├── merge-tracks.md    # Merge parallel development tracks
│   ├── worktree-status.md # Show detailed worktree status
│   ├── worktree-list.md   # List all worktrees
│   └── worktree-cleanup.md # Clean up worktrees and branches
├── examples/              # Usage examples
│   ├── complete-workflow-example.md           # Full PRD-to-deployment walkthrough
│   ├── individual-agent-usage.md             # 10 targeted scenarios
│   ├── multi-language-examples.md            # Multi-language project examples
│   └── parallel-tracks-example.md            # Parallel development demonstration
└── docs/                  # Work directories and archives
    ├── development/       # Archived development history
    ├── adrs/              # Architecture Decision Records (generated)
    ├── api/               # API documentation (generated)
    ├── features/          # Feature specifications (generated)
    ├── planning/          # PRDs and tasks (generated)
    ├── reviews/           # Code reviews (generated)
    └── sprints/           # Sprint definitions (generated)

Cost Analysis

Plugin Development

Manual Development (75 agents): 250-350 hours × $150/hour = $37,500-52,500
AI-Assisted Build: ~6 hours × $0.50/hour = $3.00
Savings: 99.99%

Using the Plugin (Per Project)

Cost per sprint optimized with T1/T2 system:

T1/T2 system ensures only relevant language agents are used
Most projects use 1-2 languages, not all 7
Shared design agents (Sonnet) work across all languages

Estimated project costs (using Sonnet/Haiku pricing):

Small project (1 sprint, 1 language): ~$0.50-0.80
Medium project (3 sprints, 2 languages): ~$4-8
Large project (10 sprints, 3 languages): ~$20-35
Enterprise polyglot (20 sprints, 5+ languages): ~$60-100

Multi-language example breakdown:

Database design (Sonnet): $0.30
Database implementation Java (T1/T2): $0.25
API design (Sonnet): $0.40
API implementation Java (T1/T2): $0.35
Frontend implementation (T1/T2): $0.40
Runtime verification (Sonnet): $0.20
Testing + Security (Sonnet): $0.50
Code reviews (Sonnet): $0.25
Total per sprint: ~$2.65

Compared to:

Human developers: 99%+ savings
Previous Opus-heavy approach: Significantly more cost-effective
Single-language AI systems: Same cost, 7x language coverage

Why This System?

Multi-Language Development Support

7 Programming Languages: Python, TypeScript, Java, C#, Go, Ruby, PHP
10+ Development Stacks: From MEAN/MERN to Spring Boot to Rails
Consistent Quality: Same T1/T2 optimization and quality gates across all languages
Language-Specific Expertise: Dedicated code reviewers for each language

Complete Development Lifecycle

Planning: PRD generation, task breakdown, sprint planning
Implementation: Database, backend, frontend, mobile
DevOps: Containerization, CI/CD, infrastructure as code
Quality: Security audits, testing, code review, documentation

Additional Features

Mobile Development: Native iOS (Swift) and Android (Kotlin)
DevOps Automation: Docker, GitHub Actions, GitLab CI, Jenkins
Cloud Infrastructure: Terraform for AWS, Azure, GCP
Scripting: Shell and Python automation scripts

Cost-Optimized Intelligence

75 Agents: Specialized for every language and task
2-Tier System: Haiku (fast/cheap) → Sonnet (balanced)
Smart Escalation: T1 handles 70-80% of work, T2 only when needed
Significant Cost Savings: Optimized model selection for each agent type

Quality Without Compromise

100% Criteria Satisfaction: Requirements validator enforces every acceptance criteria
Runtime Verification: Applications must launch without errors
Automated Testing: All tests must pass (100% pass rate required)
Security First: OWASP Top 10 audits on all implementations
80%+ Test Coverage: Automated test generation
Language-Specific Review: Code reviewers for Python, TypeScript, Java, C#, Go, Ruby, PHP
Manual Testing Documentation: Comprehensive guides for human verification

Use Cases

Greenfield Projects

Build full-stack applications from scratch in any supported language
Generate PRD, design database and APIs, implement frontend/backend
Deploy with Docker and CI/CD pipelines

Multi-Language Enterprises

Microservices architecture with different languages per service
Unified quality standards across Python, Java, Go, C# codebases
Consistent code review and security practices

Mobile + Backend

Native iOS/Android apps with backend APIs
Choose optimal backend language (Python for ML, Go for performance, etc.)
Automated API client generation and integration

DevOps & Infrastructure

Containerize existing applications with Docker
Create CI/CD pipelines for automated testing and deployment
Provision cloud infrastructure with Terraform

Legacy Modernization

Migrate from one language/framework to another
Refactor monoliths to microservices
Add tests and security audits to existing code

Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

Potential Contributions:

Additional language support (Rust, Scala, Kotlin backend, etc.)
New agent categories (testing frameworks, monitoring, etc.)
Language-specific example projects
Performance optimizations and best practices

License

MIT License - See LICENSE file for details.

Development History

This plugin was developed through an iterative AI-assisted process. Development history and architectural decisions are archived in docs/development/ for reference.

Evolution:

v1.0: Initial release with 27 agents (Python/TypeScript only)
v2.0: Expanded to 66 agents with 7 languages, mobile, DevOps, infrastructure
v2.1: Expanded to 73 agents with language-specific performance auditing, progress tracking, and parallel development tracks
v2.3: Added PR-based workflow, comprehensive sprint summaries, and git worktree support for parallel tracks
v2.4: Added runtime verification with automated testing and manual testing documentation (75 agents)
v2.5: Added workflow compliance agent and worktree management commands (76 agents, 10 commands total)

Support

Issues: https://github.com/michael-harris/claude-code-multi-agent-dev-system/issues
Documentation: See examples/ directory
Installation Help: See installation section above
Feature Requests: Open an issue with the enhancement label

Frequently Asked Questions

Q: Do I need to use all 76 agents? A: No. The system automatically selects relevant agents based on your project's language and requirements. Most projects use 10-15 agents.

Q: Can I mix languages in one project? A: Yes. The system excels at polyglot projects. For example, Go backend + TypeScript frontend + Python ML services.

Q: What if my language isn't supported? A: The generic Python agents can handle similar languages (like Ruby for scripts), or you can contribute new agents following the existing patterns.

Q: How does cost optimization work across languages? A: All languages use the same T1/T2 system. Java, C#, Go, Ruby, PHP developers benefit from the same 60-70% cost savings as Python/TypeScript.

Q: Can I use this for existing projects? A: Yes. You can use individual agents for specific tasks (add tests, create Dockerfile, generate API endpoints, etc.) without the full workflow.

Q: What quality standards do the mobile agents follow? A: They follow the same quality standards as web development: code review, security audits, testing, and iterative refinement with T1/T2 escalation.

Built with Claude Code - Demonstrating multi-agent AI development systems.

Name		Name	Last commit message	Last commit date
Latest commit History 57 Commits
.claude-plugin		.claude-plugin
agents		agents
commands		commands
docs		docs
examples		examples
.gitignore		.gitignore
ENHANCEMENTS_V2.1.md		ENHANCEMENTS_V2.1.md
ENHANCEMENTS_V2.5.md		ENHANCEMENTS_V2.5.md
LICENSE		LICENSE
README.md		README.md
install-local.sh		install-local.sh
plugin.json		plugin.json

License

michael-harris/claude-code-multi-agent-dev-system

Folders and files

Latest commit

History

Repository files navigation

Multi-Agent Development System

76 Specialized Agents

Cost Optimization (T1/T2 System)

Quality Gates

Supported Languages & Frameworks

Development Stacks Supported

New Capabilities

Installation

From GitHub (Recommended)

From Local Path (Development)

Verify Installation

Quick Start

Full Workflow

Individual Agent Usage

Progress Tracking & Parallel Development

Progress Tracking (Resume Functionality)

Parallel Development Tracks

Worktree Management Commands

Development History Preservation

Architecture

Hierarchical Orchestration

T1→T2 Escalation Flow

Quality Gate System

Model Distribution

Examples

Example 1: Complete Workflow

Example 2: Individual Agent Usage

Example 3: Multi-Language Projects

Example 4: Parallel Development Tracks

Agent Reference

Planning (3 agents)

Orchestration (5 agents)

Database (15 agents)

Backend (22 agents)

Frontend (4 agents)

Python (2 agents)

Quality (11 agents)

Scripting (4 agents) - NEW

DevOps (4 agents) - NEW

Infrastructure (2 agents) - NEW

Mobile (4 agents) - NEW

Repository Structure

Cost Analysis

Plugin Development

Using the Plugin (Per Project)

Why This System?

Multi-Language Development Support

Complete Development Lifecycle

Additional Features

Cost-Optimized Intelligence

Quality Without Compromise

Use Cases

Contributing

License

Development History

Support

Frequently Asked Questions

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Contributors 2

Uh oh!

Languages

Packages