Sample Claude Code Agent Team

Claude Code Multi-Agent Development Sample

A sample configuration for multi-agent development workflows using Claude Code. It shows how to set up a team of specialized AI agents that collaborate through a spec-driven process: a Full Stack lead orchestrates three specialists — Coding (application development), DevOps (infrastructure/deployment), and Review (code quality/security) — backed by Skills, Rules, and Plugins for the tooling and knowledge needed to ship a full-stack application to AWS.
Disclaimer: This repository is provided as an example only and is NOT approved for production use. The agent configurations, rules, and workflows are starting points — not production-ready defaults. You should review, adjust, and tailor them to fit your own project requirements, team conventions, and security posture. Adoption of this sample requires organizational legal review — you must complete the LLM Legal Approval table before use.

Overview

This repo provides a sample .claude configuration with five core agents that work together:

¹ From fullstack-agent's frontmatter, which applies only if it is spawned as a subagent. When the main agent assumes the role (the normal case), the session's own model and effort level apply — set them with /model and /effort.

Additional on-demand agents:

How It Works

1. Plan — the main agent assumes the role of fullstack-agent, researches the problem, writes a spec (spec.md, design.md), and creates a wide, file-disjoint task plan (tasks.md) sized so a pool of same-role agents can each claim a task at once
1a. Hand off — planning stops there with a Handoff Package (spawn plan, TaskCreate payloads, group order, open risks) that the main agent then executes as lead. The role runs in the main agent rather than as a subagent because a subagent…