A full product build pipeline powered by specialized Claude Code subagents. Go from a business idea to a production-deployed application with human-in-the-loop checkpoints at every critical decision point.
This is a collection of specialized Claude Code subagents organized into a complete product development pipeline. The master-orchestrator coordinates everything — you just describe your idea and approve the checkpoints.
# 1. Clone this repo
git clone https://github.com/yourusername/buildsquad-claude-code-agents
# 2. Install agents to Claude Code
./install.sh
# 3. Open Claude Code in your project directory and say:
# "Build me a product" or "I have a new idea"The pipeline has 7 phases with 6 human review checkpoints:
Business Idea
│
▼
┌─────────────────┐ CHECKPOINT 1: PRD Approval
│ Product Phase │ ──────────────────────────────► Human reviews market
│ • Market Research research + PRD research & PRD
│ • PRD Writing │
└─────────────────┘
│
▼
┌─────────────────┐ CHECKPOINT 2: Architecture Approval
│ Architecture │ ──────────────────────────────► Human reviews system
│ • System Design │ design decisions
└─────────────────┘
│
▼
┌─────────────────┐ CHECKPOINT 3: Ticket Review
│ Planning Phase │ ──────────────────────────────► Human reviews GitHub
│ • GitHub Issues │ issues & sprint plan
└─────────────────┘
│
▼
┌─────────────────┐ CHECKPOINT 4: Code Review
│ Engineering │ ──────────────────────────────► Human reviews code,
│ • Development │ tests, security
│ • Code Review │
│ • Testing │
└─────────────────┘
│
▼
┌─────────────────┐ CHECKPOINT 5: Security Sign-off
│ Security Audit │ ──────────────────────────────► Human approves security
└─────────────────┘ clearance
│
▼
┌─────────────────┐ CHECKPOINT 6: Production Deploy Approval
│ Deployment │ ──────────────────────────────► Human approves
│ • Staging │ production push
│ • Production │
└─────────────────┘
│
▼
┌─────────────────┐
│ Monitoring │
│ • Error Tracking│
│ • Alerting │
│ • Runbook │
└─────────────────┘
│
▼
Production App
At each checkpoint, the orchestrator writes a clear summary of what was built, what needs review, and the exact command to type to continue.
The pipeline saves state to .buildsquad/project-state.json. If interrupted:
resume from architecture ← continue from architecture phase
resume from engineering ← continue from engineering phase
resume from deployment ← continue from deployment phase
| Agent | Description | Model |
|---|---|---|
| master-orchestrator | Top-level pipeline controller. Start here for new products. | Opus |
| product-orchestrator | Coordinates product discovery phase (market research → PRD) | Sonnet |
| engineering-orchestrator | Coordinates full engineering lifecycle (dev → review → security) | Sonnet |
| Agent | Description | Model |
|---|---|---|
| market-research-agent | Market sizing, competitor analysis, customer pain points | Sonnet |
| product-manager-agent | Writes PRDs with user stories and acceptance criteria | Opus |
| product-strategy-advisor | Build/kill/enhance decisions and feature prioritization | Sonnet |
| github-project-manager-agent | Creates GitHub Issues, milestones, and project boards from PRD | Sonnet |
| technical-writer-agent | Writes README, API docs, deployment guides, architecture docs | Sonnet |
| Agent | Description | Model |
|---|---|---|
| system-architect | Designs scalable, maintainable system architectures | Sonnet |
| staff-engineer | Senior-level implementation with production-quality code | Sonnet |
| code-refactorer | Improves code quality, readability, and maintainability | Sonnet |
| premium-ux-designer | High-end UI/UX design and user flow optimization | Sonnet |
| git-commit-helper | Conventional Commits message formatting | Sonnet |
| shadcn-component-researcher | shadcn/ui component research and usage patterns | Sonnet |
| shadcn-implementation-builder | Builds production-ready shadcn/ui components | Sonnet |
| shadcn-quick-helper | Quick shadcn/ui installation and usage guidance | Sonnet |
| shadcn-requirements-analyzer | Breaks complex UI features into shadcn component specs | Sonnet |
| Agent | Description | Model |
|---|---|---|
| code-reviewer | Comprehensive code review for quality and correctness | Sonnet |
| senior-code-reviewer | Staff-level review: architecture, security, production readiness | Sonnet |
| qa-test-engineer-agent | Unit, integration, and E2E tests with 80%+ coverage target | Sonnet |
| security-auditor-agent | OWASP Top 10 audit, secrets scan, dependency vulnerabilities | Opus |
| performance-agent | N+1 queries, bundle size, caching gaps, algorithmic issues | Sonnet |
| Agent | Description | Model |
|---|---|---|
| devops-infrastructure-agent | CI/CD pipelines, Docker, deployment to staging and production | Sonnet |
| monitoring-observability-agent | Error tracking, logging, health checks, alerting, runbook | Sonnet |
When you run the pipeline, a .buildsquad/ directory is created in your project:
.buildsquad/
├── project-state.json ← master state (phase, approvals, decisions)
├── product/ ← market research, PRD, user stories
├── architecture/ ← system design, ADRs, approved architecture
├── planning/ ← ticket map, sprint plan
├── engineering/ ← progress tracker
├── quality/ ← test, security, and performance reports
├── devops/ ← deployment configs, runbook, secrets guide
└── checkpoints/ ← human review files (the gates)
This directory is the shared memory for all agents. It makes the pipeline resumable and auditable.
The templates/ directory contains:
CLAUDE.md.template— copy to your project asCLAUDE.mdto give all agents consistent instructionsproject-state.json.template— the project state schema.buildsquad-structure/— the expected directory structure with documentation
./install.sh # Install all agents
./install.sh --dry-run # Preview without making changes
./install.sh --remove # Remove all installed symlinksAgents are symlinked to ~/.claude/agents/ so Claude Code can discover them automatically.
Have a useful agent to share? Contributions welcome. Submit a pull request with your agent configuration following the existing format.
Agent file format:
---
name: your-agent-name
description: When to use this agent. Include <example> blocks.
model: sonnet # or opus, haiku
color: green # cyan, purple, pink, green, yellow, blue, red, orange
---
Your agent's system prompt in markdown...Watch the tutorial: Learn how to use and create subagents in this YouTube video.