Three Man Team

Three Man Team is an open-source, MIT-licensed framework for running a structured three-agent AI development workflow inside Claude Code. Built by Russell Aaron from production use on a real SaaS platform, it ships as a cloneable GitHub repository that installs directly into your project or globally across all projects.

What It Is

Three Man Team is a prompt-engineering and agent-workflow framework that assigns three distinct roles to AI agents: Architect (plans, deploys, orchestrates), Builder (builds exactly what the brief specifies), and Reviewer (catches what the Builder missed). All three run inside a single Claude Code session — Arch is the main agent, and it spins up Bob and Richard as subagents via Claude Code's Agent tool. There are no separate windows or sessions to manage.

Why Three Agents

The framework's README argues that solo agents drift with no one to catch wrong turns, while large teams generate coordination overhead that erodes productivity gains. Three is described as the minimum for meaningful review and the maximum before the team starts managing itself instead of the work. The roles map to how real software ships: one agent owns the system and the deploy, one builds fast and clean, and one enforces quality before anything ships.

Token Optimization Architecture

Every install includes a token-optimizer skill that auto-loads via CLAUDE.md. Five rules are baked in at session start:

• Trust skills and memory — skip redundant file reads
• Kill speculative tool calls
• Parallelize calls that can run in parallel
• Route long output to a subagent
• Delete restatements of what the user said

The README also recommends pairing Three Man Team with RTK, a separate bash-layer tool that compresses find, ls, and grep output before it reaches Claude's context window, for compounding savings.

Setup Path

Installation is a single git clone into .claude/skills/three-man-team followed by running ./setup. The setup script generates a manifest.md — a single source of truth for team names, role filenames, handoff directory, repo, and branch. Per-project and global install paths are both supported. Templates ship with named personas (Arch, Bob, Richard) ready to use, or a blank-slate generic template with [CUSTOMIZE] placeholders for building custom personas from scratch.

Update: v1.3.0 — The Manifest

Version 1.3.0, published June 2026, introduced manifest.md as the canonical install record, replacing the retired VERSION file. The version check system was also reworked to read a releases/latest.json registry instead of hitting the GitHub API directly. Critical updates are flagged as mandatory checkpoints; non-critical updates are optional. Arch reads actual project files before suggesting any changes, and nothing applies without user confirmation.

Compatibility and Audience

The framework targets developers using Claude Code, VS Code, Cursor, or any AI tool that supports context files. A "Three Man Team Pro" tier with pre-built teams for developers, marketers, and content creators is listed as a waitlist product on the developer's site, but the core framework is free and open source under MIT.