Agentic AI Atlas

II.

Specialization reference

specialization:automaker

Reading · 3 min

Automaker reference

AutoMaker is an autonomous AI development studio that transforms software building: describe features on a Kanban board and AI agents implement them using Claude Agent SDK. It features git worktree isolation for safe parallel development, streaming UI updates, multi-agent task execution, and integrated test automation with Vitest and Playwright.

Specializationwiki/library/automaker.mdOutgoing · 1Incoming · 2

AutoMaker

**Source:** AutoMaker by AutoMaker-Org **Category:** Autonomous AI Development Studio **Priority:** High

Overview

Unlike traditional development tools, AutoMaker positions AI agents as autonomous implementers that work in isolated branches, write code, run tests, and submit changes for human review before merge. The Kanban board metaphor provides intuitive project management while agents handle implementation details.

Key Features

**Kanban-Based Workflow** with feature cards moving through Backlog, Ready, In Progress, Review, Done
**Git Worktree Isolation** ensuring agents work safely in separate branches
**Multi-Agent Parallel Execution** with configurable concurrency limits
**Streaming Progress Updates** showing real-time agent activity
**Integrated Test Automation** with Vitest (unit) and Playwright (E2E)
**Code Review Gates** with configurable approval policies
**Convergence Loops** for automatic test failure resolution
**Build & Deploy Pipeline** with release notes generation

Process Files

File	Description	Primary Agents
`automaker-orchestrator.js`	Full lifecycle (all 5 phases)	All agents
`automaker-feature-pipeline.js`	Feature intake, decomposition, dispatch	Feature Planner
`automaker-agent-execution.js`	Single-feature execution with worktree	Code Generator, Test Runner, Worktree Manager
`automaker-review-ship.js`	Review, merge, deploy, release	Code Reviewer, Deployment Engineer

Agent Personas

Agent	Role	ID
Feature Planner	Feature Decomposition & Planning	`automaker-feature-planner`
Code Generator	Implementation & Fixes	`automaker-code-generator`
Test Runner	Test Execution & Analysis	`automaker-test-runner`
Code Reviewer	Quality Review & Gates	`automaker-code-reviewer`
Worktree Manager	Branch Isolation & Merges	`automaker-worktree-manager`
Progress Streamer	Real-time UI Updates	`automaker-progress-streamer`
Deployment Engineer	Build, Deploy & Release	`automaker-deployment-engineer`

Skills

Skill	Description	Agent
feature-intake	Parse features from text, images, screenshots	Feature Planner
agent-dispatch	Batch and dispatch features to execution agents	Feature Planner
worktree-isolation	Git worktree lifecycle management	Worktree Manager
streaming-progress	Real-time progress streaming to UI	Progress Streamer
test-automation	Vitest and Playwright test execution	Test Runner
code-review-gate	Code review with quality thresholds	Code Reviewer
kanban-management	Kanban board state management	Feature Planner

Tech Stack (from source project)

**Frontend:** React, Vite, Electron (desktop app)
**Backend:** Express.js server
**Testing:** Playwright (E2E), Vitest (unit)
**Monorepo** with shared packages
**Claude Agent SDK** for agent orchestration

Usage

bash

# Full lifecycle - orchestrate features from intake to deployment
babysitter run:create --process methodologies/automaker/automaker-orchestrator --input examples/multi-feature-sprint.json

# Feature pipeline only - decompose and prioritize features
babysitter run:create --process methodologies/automaker/automaker-feature-pipeline --input examples/feature-batch.json

# Single feature execution - implement one feature in isolation
babysitter run:create --process methodologies/automaker/automaker-agent-execution --input examples/single-feature.json

# Review and ship - review, merge, and deploy completed features
babysitter run:create --process methodologies/automaker/automaker-review-ship --input examples/review-and-deploy.json

Automaker reference