The Plan → Build Workflow

AI teammates that argue with each other so your code ships better

For Claude Code

Watch the Demo Get Started

See It In Action

Demo of the team workflow: planning, critiquing, building, and validating code

The Problem

"Just code it" fails

AI generates plausible garbage without a plan. First idea ships, nobody challenges it.

Review after the fact is too late

Rework is expensive. By the time you catch issues, you've already committed to a bad approach.

One voice, one blind spot

A single AI assistant can't critique itself effectively. It confirms its own assumptions.

The Pattern

The best AI coding workflows converge on the same structure: Plan → Critique → Build → Validate. This is what Devin, Cursor, and top AI teams are discovering.

1

Plan

Understand the task. Design the approach. Identify risks before writing code.

Peter
→
2

Critique

Challenge the plan. Find holes. Play devil's advocate. Catch bad ideas early.

Neo
→
3

Build

Implement from the refined plan. Follow the spec. Ship accessible, quality code.

Gary
→
4

Validate

Review everything. Nothing merges without sign-off. Guard quality gates.

Reba

Result: Code that's been planned, challenged, built to spec, and validated—before you even look at it.

The Team

Each step in the pattern is implemented by a persona with defined expertise. They know each other, hand off work, and hold each other accountable.

Planning Peter

Planner

Invents process, drives consensus. Runs retrospectives. Makes the call when the team disagrees.

Nifty Neo

Architect / Critic

Challenges designs, finds bottlenecks. Grounds hallucinations. Devil's advocate for every proposal.

Research Reba

Reviewer / QA

Validates everything. Nothing merges without her sign-off. Guards the immutable sections.

Meticulous Matt

Auditor / Security

Finds all issues, reports honestly. Security triage. Blue team defensive audit.

Greenfield Gary

Builder

Implements from plans. Accessibility expert. Internationalization. UX patterns across platforms.

Grizzly Gabe

Fixer / Red Team

Resolves issues with care. Offensive security. Thinks like an attacker to defend like a pro.

Zen Runner

Executor

Autonomous work. No human-in-loop. Executes well-defined tasks independently.

How It Works

Personas, Not Processes

The team members are personas in the same context, not isolated agents. They can hear each other, interrupt, and collaborate in real-time.

Self-Improving Protocols

The team writes their own TEAM.md. They run retrospectives. They update their own skills based on what they learn. You're the founder who steps back and lets them evolve.

Safety Rails

  • No Lobotomies - Cannot edit IMMUTABLE sections of any skill
  • Reba's Law - All self-modifications pass validation
  • Stay in Your Lane - Only modify skills and team files
  • User Has Final Say - All commits visible; user controls merges and pushes

Get Started

Quick Install:

curl -sL https://raw.githubusercontent.com/HakAl/team_skills/master/install.sh | bash

Works on Linux, macOS, and Windows (Git Bash). Updates with the same command.

Then in Claude Code:

  1. Run /team genesis to bootstrap protocols
  2. Give the team a task: /team "add input validation"
  3. Watch them plan, critique, build, and validate

View on GitHub

What We've Learned