Agent Patterns

Agent Patterns

AI agent patterns monorepo demonstrating structured agentic workflows for intelligent automation.

Overview

This repository implements a **structured agentic workflow system** that combines human planning with AI execution. It demonstrates various agent patterns including sequential processing, routing, parallel processing, orchestrator-worker, and evaluator-optimizer patterns.

Architecture

• Backend: Nest.js API with streaming responses using Vercel AI SDK
• Frontend: React + Vite + Material UI with real-time streaming UI
• Monorepo: npm workspaces managing packages/api and packages/webapp

Workflow System

This project uses a unique **issue-driven development workflow** optimized for AI collaboration:

📋 Issue Management Structure

issues/           # Requirements specifications
├── 0-bootstrap-plan.md
├── 1-project-setup.md
└── 2-backend-implementation.md

docs/             # Detailed implementation plans
├── plan_0-bootstrap-plan.md
├── plan_1-project-setup.md
└── plan_2-backend-implementation.md

scripts/          # Automation tooling
├── create-issue.sh      # Generate new issues
├── complete-task.sh     # Mark tasks complete
└── claude-context.sh    # AI agent orchestration

todo.md           # Central task tracker

🔄 Workflow Process

1. **Issue Creation**

# Create new issue with templates
npm run issue "Feature Name"

• Auto-generates issue ID and file structure
• Creates both requirement spec (issues/) and implementation plan (docs/)
• Updates central task tracker (todo.md)

2. **Bootstrap from Master Plan**

When you have a comprehensive master plan (like `plan-eval.md`), use the bootstrap approach:

# Create bootstrap issue
npm run issue "Bootstrap [Feature] from Master Plan"

Edit the created issue to instruct Claude to decompose your master plan:

# Issue X: Bootstrap [Feature] from Master Plan

## Requirement
Read the `[master-plan].md` fi