</>MCP Agents Market
Agent

three-man-team

by russelleNVy948ShellUpdated 2026-06-09

A structured 3-agent AI dev team — Architect, Builder, Reviewer. Built from production use. Token-optimized. Works with Claude Code, VS Code, Cursor, and any AI that supports context files.

Claude CodeVS CodeCursor

Three Man Team is a structured three-agent development workflow for AI coding assistants that assigns distinct roles—Architect, Builder, and Reviewer—to prevent common AI coding failures like scope drift and excessive token usage. Built from production experience, it implements clear handoffs and review gates to ensure AI agents plan, build, and verify work systematically. The system runs within a single Claude Code session using sub-agent spawning, with the Architect orchestrating the Builder and Reviewer as needed. It includes token optimization rules, auto-update checks, and works with Claude Code, VS Code, Cursor, and any AI client that supports context files.

Key Features

Three-role agent structure (Architect for planning/deployment, Builder for implementation, Reviewer for quality control) with defined handoffs
Single-session workflow where Arch spawns Bob and Richard as sub-agents via Claude Code's Agent tool
Token optimization discipline baked into CLAUDE.md with five core rules to prevent redundant file reads and speculative tool calls
Auto-update system that checks releases/latest.json and walks through applicable changes based on actual installed files
Per-project or global installation modes with setup script that generates manifest.md as single source of truth
Pre-built persona templates (Arch, Bob, Richard) with customizable 'Who You Are' sections and flexible renaming
Workflow discipline that prevents common AI coding failures: reading entire codebases unnecessarily, adding unsolicited features, and mid-task drift
MIT licensed and built from real production SaaS development experience

Use Cases

  • 01Structured AI-assisted feature development with automatic planning, building, and review phases
  • 02Token-efficient AI coding sessions that avoid redundant context loading and speculative work
  • 03Multi-agent code review workflows where one agent builds and another independently verifies
  • 04Team-based AI development processes for solo developers who want systematic quality gates
  • 05Project workflows requiring clear separation between architecture decisions, implementation, and verification
  • 06AI coding projects where controlling scope creep and maintaining discipline is critical

Related Agents

View more

three-man-team — FAQ

What is three-man-team and what does it do?+

Three-man-team is a structured AI sub-agent system that divides development work among three specialized roles: an Architect who plans and deploys, a Builder who implements exactly to spec, and a Reviewer who validates the work. It prevents common AI coding failures like scope drift, excessive token use, and mid-task wandering by enforcing clear handoffs and review gates.

How do I install three-man-team in my project?+

For per-project installation, clone the repository into your project's .claude/skills/three-man-team directory, then run the ./setup script which will guide you through the process and provide exact commands and prompts. For global installation, clone to ~/.claude/skills/three-man-team, run setup once, then copy template files into each project where you want to use it.

Which AI clients and editors work with three-man-team?+

Three-man-team works with Claude Code, VS Code, Cursor, and any AI coding assistant that supports context files and sub-agent spawning. The primary workflow runs in a single Claude Code session where the Architect agent spawns the Builder and Reviewer as sub-agents.

Do I need API keys or credentials to use three-man-team?+

No additional API keys are required beyond what your AI coding client already uses. The agent system runs within your existing Claude Code or compatible client session and relies on that client's authentication.

Is three-man-team free to use?+

Yes, three-man-team is MIT licensed and free forever for any use. The repository is open-source and there is no cost to download, install, or use it in your projects.

How does three-man-team optimize token usage?+

It includes a token-optimizer skill that enforces five rules: trusting skills/memory instead of re-reading files, killing speculative tool calls, parallelizing when possible, routing large outputs to sub-agents, and deleting redundant restatements. These rules are automatically loaded via CLAUDE.md in every session.

How do I install three-man-team?+

Open the source repository on GitHub and follow its README. three-man-team is a agent — MCP Agents Market links you directly to the official repo.

Is three-man-team free?+

three-man-team is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches