Reference guide
Plain English glossary
Every term you'll encounter while building with AI — explained without jargon.
Claude Code
An AI assistant that lives in your computer's command-line window and works directly inside your project files — reading, writing, and editing code on your behalf.
Like a contractor who works inside your actual office, not one who sends emails from across town.
CLAUDE.md
A plain text document you write once that tells the AI about your project — your stack, your rules, your current focus. It loads automatically at the start of every session.
The onboarding document you'd hand a new hire on day one.
Global vs project file
You have two instruction files. The global one applies to every project — your personal defaults. The project one only applies to that specific product.
Your personal work style is global. The brief for a specific client is project-level.
Context window
The amount of information the AI can hold in working memory during one session. The bigger the window, the more files and history it can track at once.
The size of someone's desk — bigger desk, more files spread out at once.
Skill
A folder with a plain-text instruction file inside. It teaches the AI a specific task — your exact process, your format, your rules. Claude loads it automatically when relevant.
A laminated instruction card above a workstation — your exact recipe, not a generic one.
Community skill
A pre-built skill someone else wrote and shared on GitHub. You download it and immediately have that capability — document generation, SEO analysis, slide creation.
A browser extension — install once, get the capability immediately.
MCP (Model Context Protocol)
The standard way to connect your AI to outside systems — your database, your files, GitHub, Slack. Once connected, the AI can look things up and take action instead of only knowing what you type.
Giving your contractor a keycard to the filing room and company database instead of them working from what you bring into the meeting.
MCP server
A specific connection to one external system. One for your database, one for your files, one for GitHub. Each is a separate tool the AI can pick up and use.
Each key on a keyring gives access to one room. More MCP servers = more rooms the AI can enter.
Subagent
A separate AI instance spawned to handle one specific job. Multiple subagents can work simultaneously on different parts of a task, then report back with results.
Dispatching specialists from your team to work in parallel, then reconvening to combine results.
Orchestrator
The main AI session that manages the overall task — breaks it into pieces, assigns work to specialists, and pulls everything back together.
The project manager who delegates to specialists, then assembles the final deliverable.
Always-on agent
An agent that runs automatically — triggered by an event (a file being saved, a PR being opened) or on a schedule (every morning at 8am). You don't trigger it. It just runs.
A security guard who patrols automatically — you don't have to ask them to check the doors.
Hook
A rule that fires automatically after every action the AI takes. For example: "after every file edit, run a quality check." The AI sees the result and fixes anything wrong before you look.
A spell-checker that runs automatically every time you finish typing — you never manually ask it to check.
Database
Where your app stores and retrieves information — user accounts, records, uploaded data. Everything your app needs to remember lives here.
The filing system behind the counter. The front of the shop is what users see; the database is everything behind it.
Deployment platform
The service that takes your code and makes it live on the internet. Every time you push an update, the platform rebuilds and publishes your app automatically.
The printer and delivery service — you write the document, it prints and ships to anyone who visits your URL.
Version control (GitHub)
Where your code lives online. Every change is saved with a description of what changed, so you can always go back to an earlier version or share code with others.
Google Docs version history for your entire codebase, plus a shared drive for your team.
API route
A URL on your server that your app can call to get data or trigger an action. When your app needs AI logic, it calls an API route on your own server, which then calls the AI.
An internal phone extension — dial the right number and a specific department handles your request.
Environment variable
A secret stored on your computer (not in your code files) that your app can read — like API keys and passwords. Keeps sensitive information out of your code.
Your bank PIN on a Post-it on your own desk rather than written inside a shared document.
Vibe coding
Building software by describing what you want in plain English and letting AI write the code — without needing to know how to code yourself.
Being the architect who draws the vision while the AI acts as the construction crew that builds it.
Fallback
What your app does when something fails — like the AI being temporarily unavailable. A good fallback queues the task for later or shows a "processing" message instead of breaking.
When your first supplier is out of stock, you automatically switch to your backup instead of cancelling the order.
Migration
A saved set of changes to your database structure — adding a new field, creating a new table. Migrations track changes so they can be applied or reversed precisely.
A documented change order for your filing system — recorded so you can undo it if needed.
Prompt injection
A security risk where hidden instructions inside a file try to trick the AI into doing something you didn't ask for — like overriding your rules or sending data elsewhere.
Someone hiding a note inside a document that says "ignore the contract and do this instead."
Stack block
A reusable section of your instruction file that describes a specific set of technologies you use. Write it once, paste it into any project that uses the same tools.
A template clause in a contract — written once carefully, reused whenever it applies.
.cursorrules
The equivalent of CLAUDE.md but for Cursor's built-in AI agent. Same idea — a plain text file at your project root that tells the AI your conventions and rules.
Same onboarding document, different contractor reading it.
Personal toolkit repo
A private GitHub repository where you store all your reusable AI assets — your global instruction file, stack blocks, skills, and agent definitions. Pull from it to start any new project instantly.
Your personal toolbox that you bring to every job site, instead of buying new tools each time.