Skip to main content

Getting Started with Babysitter

Welcome to Babysitter! This guide will help you go from zero to running your first AI-orchestrated development workflow in just a few minutes.


30-Second Overview

Babysitter = Claude Code + autopilot + quality control

Instead of:

You: "Build me a login page"
Claude: *builds something*
You: "Make it better"
Claude: *tweaks it*
You: "Run the tests"
Claude: *tests fail*
You: "Fix the tests"
... (repeat for hours)

You just say:

/babysitter:call build a login page with 90% quality target

Babysitter handles the iteration, testing, and quality checks automatically. Come back later, it's still working (or waiting for your approval).


Choose Your Path

If you are...Start here
Impatient (just want to try it)Quickstart - 10 minutes to first run
Thorough (want to understand first)Keep reading this page, then Installation
Already installedFirst Run Deep Dive to understand what happened
Coming backJump to the Tutorials for deeper projects

What is Babysitter?

Babysitter is an orchestration framework for Claude Code that transforms how you work with AI-assisted development. Instead of manually iterating with Claude until your code is "good enough," Babysitter automates the entire process with:

  • Automatic quality convergence - Set a quality target (like 85%), and Babysitter iterates until it's achieved
  • Session persistence - Close your laptop, come back tomorrow, and pick up exactly where you left off
  • Human-in-the-loop approval - Add approval gates (breakpoints) for critical decisions before they execute
  • Complete audit trails - Every decision, iteration, and change is recorded in an event journal

Think of Babysitter as a project manager for your AI coding sessions - it keeps track of progress, ensures quality standards are met, and never loses context.

Why You Will Love Babysitter

Without BabysitterWith Babysitter
"Claude, can you improve that?" (repeat 10x)Set quality target once, iterate automatically
Lose all context when session endsResume from any point, even days later
Hope the AI made good decisionsReview and approve at key decision points
No record of what happenedComplete journal of every action
Run tasks one at a timeParallel execution for faster results

Key Benefits

1. Quality Convergence

Stop manually asking Claude to "make it better." Define your quality target (test coverage, code standards, etc.) and Babysitter iterates until it's achieved.

# Example: Iterate until 85% quality score
claude "/babysitter:call implement user auth with TDD, 85% quality target"

2. Never Lose Progress

Every action is recorded in an event-sourced journal. Session interrupted? No problem.

# Resume exactly where you left off
claude "Resume the babysitter run for the auth feature"

3. Human-in-the-Loop Control

Add approval gates for critical decisions. Review context, approve or reject, and only then does execution continue.

# Babysitter will pause for approval before deploying
claude "/babysitter:call deploy to production with breakpoint approval"

4. Structured Workflows

Choose from built-in methodologies (TDD, Spec-Kit, GSD) or create your own. Consistent, repeatable processes across your team.


Quick Navigation

I want to...Go to...
Install BabysitterInstallation Guide
Run my first workflow (5 min)Quickstart
Understand what happenedFirst Run Deep Dive
See all commandsCLI Reference
Learn about Quality ConvergenceQuality Convergence Guide

Prerequisites

Before you begin, ensure you have the following:

Required Software

SoftwareVersionHow to CheckInstallation Guide
Node.js20.0.0+ (22.x recommended)node --versionnodejs.org
npm8.0.0+npm --versionComes with Node.js
Claude CodeLatestclaude --versionClaude Code Docs
SoftwarePurposeInstallation
nvmManage Node.js versions easilynvm-sh/nvm
GitVersion control for your projectsgit-scm.com
jqParse JSON output from CLIbrew install jq (macOS)

Knowledge Prerequisites

LevelWhat You Should Know
RequiredBasic command line usage (cd, ls, npm)
RequiredHow to use Claude Code (basic prompting)
HelpfulJavaScript/TypeScript basics (for custom processes)
OptionalTest-driven development concepts

Verify Your Environment

Run these commands to confirm you're ready:

# Check Node.js (need 20.0.0+)
node --version
# Expected: v20.x.x or v22.x.x

# Check npm
npm --version
# Expected: 8.x.x or higher

# Check Claude Code
claude --version
# Expected: Claude Code version info

If any command fails, install the missing software before continuing.


Installation Overview

Getting Babysitter running involves three steps:

  1. Install the Babysitter plugin (via Claude Code's plugin system)
  2. Install the CLI packages (npm global install for orchestration commands) Estimated time: 5-10 minutes

Ready? Head to the Installation Guide for step-by-step instructions.


Your Learning Path

Day 1: Get Started (Today!)

  1. Read this introduction (you are here)
  2. Install Babysitter (5 min)
  3. Complete the Quickstart (10 min)
  4. Understand your first run (10 min)

Week 1: Build Confidence

  1. Try different quality targets (80%, 90%, 95%)
  2. Experience session resumption (close and resume a run)
  3. Use breakpoints for approval workflows
  4. Explore the TDD methodology

Week 2+: Level Up

  1. Compare methodologies (TDD, GSD, Spec-Kit)
  2. Customize quality targets and iteration limits
  3. Learn about parallel execution
  4. Create your first custom process (advanced)

How Babysitter Works (The Big Picture)

You: "Build a todo API with TDD"
|
v
+------------------+
| Babysitter |
| Orchestration |
| |
| 1. Research |---> Analyze codebase
| 2. Plan |---> Create specifications
| 3. Implement |---> TDD iterations
| 4. Quality |---> Score and improve
| 5. Approve |---> Human checkpoint
| 6. Complete |---> Deliver result
+------------------+
|
v
Everything logged to .a5c/runs/<runId>/journal/

The Magic: Event-Sourced Persistence

Every action Babysitter takes is recorded as an event in a journal:

{"type":"RUN_STARTED","runId":"01KFFTSF8TK8C9GT3YM9QYQ6WG","timestamp":"2026-01-25T10:30:00Z"}
{"type":"TASK_STARTED","taskId":"research-001","timestamp":"2026-01-25T10:30:01Z"}
{"type":"TASK_COMPLETED","taskId":"research-001","result":{...},"timestamp":"2026-01-25T10:30:45Z"}

This means:

  • Crash recovery: Replay the journal to restore exact state
  • Audit trail: Complete history of every decision
  • Debugging: Trace exactly what happened and when

What You Will Build

In this getting started guide, you will:

Quickstart Tutorial (10 minutes)

Build a simple calculator module with TDD:

  • Write tests first
  • Implement code to pass tests
  • Achieve 80% quality score
  • See automatic iteration in action

Expected Outcome

calculator/
calculator.js # Implementation
calculator.test.js # Test suite
README.md # Generated documentation

Quality Score: 85/100
Tests: 12 passing
Coverage: 92%

Getting Help

Stuck? Try These Resources

ResourceBest For
Troubleshooting GuideInstallation issues
FAQCommon questions
GitHub IssuesBug reports
GitHub DiscussionsQuestions and ideas

Common First-Time Issues

ProblemSolution
"command not found: babysitter"Run npm install -g @a5c-ai/babysitter
Plugin not appearingRestart Claude Code after installation

Ready to Begin?

Your journey starts with installation. Let's get Babysitter running on your machine.

Next Step: Installation Guide


Quick Reference Card

Keep these commands handy:

# Start a new run
claude "/babysitter:call <your request>"

# Resume an existing run
claude "/babysitter:call resume the babysitter run"
# Check available skills
/skills

# View run history
ls .a5c/runs/

Ready to transform your AI-assisted development?

Start Installation