The Setup That Makes Claude Code Actually Work
Claude Code is powerful. Without the right configuration files, that power is inconsistent. This guide installs the instruction layer - CLAUDE.md, MEMORY.md, a /docs folder - that keeps your agent reliable across every session.
3 framework guides that map where your AI setup actually stands - and what the next step looks like.
No credit card. No spam. Files live in your account.
How to set up Claude Code
Proper Claude Code setup requires three files in your project root: CLAUDE.md (your project rules and coding standards), MEMORY.md (decisions that persist across sessions), and a /docs folder (structured project context). Without these, your agent starts every session without memory, drifts from your standards, and requires constant re-explanation. This guide provides the templates, the setup prompts, and 13 video lessons that walk through each layer.
- CLAUDE.md is read by Claude Code automatically before every response - your rules stay in place across every session without re-explaining them
- MEMORY.md stores decisions that survive session resets, model updates, and new conversations
- Skills and Hooks are the advanced layers - reusable commands and automated checks that eliminate manual supervision
What breaks when Claude Code has no setup
The result
Your output changes between conversations. Your standards drift. You become the only thing holding your Claude Code workflow together.
One file changes all of that.
CLAUDE.md is a file Claude Code reads before every response - not a prompt, but a persistent instruction layer. Set it up once and your rules survive every session reset, model update, and new conversation.
The 5-layer setup framework
The Agent Orchestration Layer (AOL) maps the five stages of Claude Code integration - from zero context to fully automated systems. This guide walks you through all five.
No setup
Claude has no context. Every session starts from zero. You re-explain the project every time.
CLAUDE.md
Claude reads your project rules before every session. Your standards stay in place without prompting.
/docs folder
Claude navigates structured project knowledge on demand. You stop explaining and start building.
Skills
Complex multi-step instructions become single-word commands. Your workflow is reproducible.
Hooks
Automated checks fire on every tool call. No supervision needed. Your rules enforce themselves.
What you learn
13 video lessons - available in your account after signing up.
Welcome to the Starter Guide
What is inside the Guide and what to do first
How to Activate Claude
The 3-step sequence that connects Claude to your project rules from the first message
The Builder's Roadmap
APL, AAL, and AOL explained - three maps that tell you where your setup stands today
CLAUDE.md - The Project Constitution
The file Claude reads before every response - it keeps your rules in place across every session
The /docs Folder
Why your agent needs structured project memory beyond a single configuration file
Advanced /docs Patterns
How to organize documentation so Claude loads only what it needs, when it needs it
Claude Skills
Write a complex instruction once and trigger it with a single word from any session
Claude Hooks
Automated rule enforcement that fires on every tool call - no manual supervision required
MEMORY.md
How to make Claude carry your decisions and preferences across every session reset
Setup Complete
Your system is active - what to build from here and how the layers compound over time
Agent Power Levels (APL)
4 levels from browser chat to multi-agent orchestration - find which stage your setup is at
Agent Access Levels (AAL)
5 levels that map how deeply your agent reaches into your data and systems
Agent Orchestration Layers (AOL)
From stateless chat to fully automated systems - the 5 layers and where you currently sit
What's inside
Seven files. Each one eliminates a specific failure mode that makes Claude Code unreliable without it.
ABA Setup Instructions
Read this first - exact prompts that configure your entire system from zero in one session
CLAUDE.md Starter
The file Claude reads before every response - your rules, in every session, without re-explaining
SKILL.md Starter
Define a complex instruction once and trigger it with a single word from any project
MEMORY.md Starter
Decisions and context that survive session resets, model updates, and new collaborators
Docs Folder (Simple)
Lightweight structure your agent reads and navigates without you explaining the project
Docs Folder (Advanced)
Full scaffolding for complex, long-running projects with multiple components
Hooks Setup Guide
Automated checks that run on every save - your agent catches errors before they compound
Clone the repo
All files live in a public GitHub repo. Clone it once and your agent can read them directly - no copy-paste, no manual setup. This is how AI builders get started.
git clone https://github.com/mrnelke/ai-builder-starter-kit
The repo contains all seven starter files in a ready-to-use structure. Clone it, point your agent at the files, and it reads the setup instructions and builds your CLAUDE.md from scratch - no manual configuration needed.
Prefer to download individual files and track your progress? Sign up above and everything is in your account.
How it works
Three steps and your instruction layer is running.
- Create an
ABAfolder in your project root and add the starter guide files there - Give your agent this prompt:
I have added the ABA Starter Guide files to the /ABA folder in this project. Please read all the files in /ABA and set up the correct project instructions for me. Create CLAUDE.md, MEMORY.md, and SKILL.md in the project root based on what we are building. Ask me any questions you need to fill them in correctly.
- Same - create an
ABAfolder and drop the MD files in - Give your agent this prompt:
I have added the ABA Starter Guide files to the /ABA folder in this project. Please read all the files in /ABA, then review the existing project and consult me on how to improve each file - CLAUDE.md, MEMORY.md, SKILL.md. For each file, tell me what you would add or change and why, then ask for my input before making changes.
Works with
3 framework guides included
Most people building with Claude Code have no clear picture of where their setup actually stands. These three guides map your current level and show exactly what the next stage requires.
Agent Power Level (APL)
Locate your current AI setup on a 4-level scale - and see exactly what becomes possible at the next level
Agent Access Level (AAL)
Map how deep your agent reaches into your system - and identify where real automation starts
Agent Orchestration Layer (AOL)
See the 5 stages of AI integration from zero context to autonomous systems - and find where you are stuck
Common questions about Claude Code setup
What is CLAUDE.md and why does Claude Code need it?
CLAUDE.md is a configuration file that Claude Code reads automatically before every session. It holds your project rules, coding standards, and workflow instructions so you do not have to re-explain them each time. Without it, every session starts without context and your agent behaves generically instead of following your specific standards.
How is Claude Code different from using ChatGPT for coding?
Claude Code operates directly inside your project - it reads your files, runs commands, edits code, and tracks changes. ChatGPT processes text you paste into a chat window. Claude Code is at APL 3, which is the inflection point where an AI agent stops talking and starts doing. That level of access requires a proper setup to work reliably across sessions.
How do you set up Claude Code for a new project?
The fastest way is to clone the ABA Starter Guide repository, add the files to an ABA folder in your project root, and run the setup prompt. Claude Code reads the setup instructions and builds your CLAUDE.md, MEMORY.md, and SKILL.md from scratch - asking you questions as it goes. The whole process takes one session.
What happens if you do not configure Claude Code before using it?
Without configuration, your agent starts every session without memory of your rules. Context you establish disappears when the conversation ends. You spend the first part of every session re-explaining the project, your output varies between sessions, and over time you become the only thing holding the workflow together.
Does this setup work with Cursor, Windsurf, and other AI coding tools?
Yes. CLAUDE.md, MEMORY.md, and the /docs structure work with any AI coding assistant that reads files from your project directory - including Cursor, Windsurf, Aider, Cline, and GitHub Copilot. The setup is file-based, not tool-specific. Claude Code is the primary tool this guide was built around, but the architecture transfers to any agent that reads your codebase.
Who this is for
Good fit
- You are using Claude Code but your agent gives different output across sessions
- You start every new project by re-explaining how you work
- Your rules disappear when a session times out or resets
- You want a proper foundation before building anything serious on top of Claude Code
Skip if
- You only use ChatGPT or Claude in the browser
- You are not building real projects
What you earn
Your first ABA credential. Confirms your instruction layer is in place and your Claude Code setup is ready to build on top of.
Where this fits
The Starter Guide is step one. Each course builds on the last.
Claude Code Starter Guide
Install it once and Claude Code stops starting from zero. Every project after that inherits your rules, your memory, and your context.
No credit card. No spam. Files live in your account.