Getting Started

This guide will help you get up and running with Brainbrew Devkit in minutes.

What is Brainbrew?

Self-correcting agent chains for Claude Code and opencode. A chain of agents takes turns. One plans, one codes, one reviews, one tests, one commits. If an agent fails, another fixes it and the chain retries — automatically.

Instead of manually deciding what agent to run next, Brainbrew uses AI-powered routing (via Haiku) to analyze each agent's output and pick the appropriate next step.

Prerequisites

• Claude Code or opencode
• Node.js 18+

Quick Setup

1. Install the Plugin

/plugin marketplace add brainbrewlabs/brainbrew-devkit
/plugin install brainbrew-devkit

After installation, restart Claude Code (or opencode) for hooks, agents, and the chain config to take effect.

2. Build a Chain (recommended)

Templates are reference examples, not best practice. Every project is different — your real workflow needs its own chain. Just describe it:

"Build me a chain for this project"
"I need: design review → implement → security scan → deploy"

The chain-builder agent reads your codebase, asks a few questions, and writes a chain that fits.

Need a specific capability? Ask skill-finder — it searches Vercel Skills, GitHub, and Anthropic's official skills, then installs a match:

"Find a skill for writing tests"
"Install a skill for database migrations"

Or start from a template (use only as a copy-from starting point):

"Set up a development workflow"

3. Start Working

/code implement login feature

The pipeline runs automatically:

• planner creates the plan
• plan-reviewer validates it
• implementer writes code
• code-reviewer checks quality (routes back if issues)
• tester runs tests (routes to debugger if fails)
• git-manager commits and creates PR

Core Concepts

Chain Flow

A chain defines the sequence of agents and routing rules:

flow:
  code-reviewer:
    routes:
      tester: "Code approved, ready for testing"
      implementer: "Code has issues, needs fixes"
    decide: |
      If code is APPROVED → "tester"
      If ANY bugs, issues → "implementer"

Memory Bus

Agents communicate via the Memory Bus:

"Tell implementer to fix the auth bug"
→ memory_add(content: "...", target: "agent:implementer")

Agent Teams

Run multiple agents in parallel at a chain step:

parallel-review:
  type: team
  teammates:
    - name: code-quality
      agent: code-reviewer
    - name: security-check
      agent: security-scan

Next Steps

• Installation — Detailed setup instructions
• Chain Workflow — Deep dive into chains
• Templates — Browse available templates