Claude Code is Anthropic's AI assistant that runs inside VS Code or your terminal, reads files, runs commands, and connects to external services through MCP. This glossary defines the 28 terms you will hit in your first week. Written in plain English, no developer background needed.
Most people who pick up Claude Code for the first time hit a wall within ten minutes. Not because it is hard, but because every explanation assumes you are a developer.
You are not. But you have heard enough about Claude Code to be curious, and you are not the type to sit still while everyone else figures it out first.
This is the reference post I wish had existed when I started.
It will not turn you into a developer. It will make sure you understand what is being talked about when people throw around terms like MCP, hooks, or subagents. And it will show you what Claude Code actually is for someone in your seat.
What is Claude Code, really?
Claude Code is a version of Claude that lives inside your code editor (VS Code) or runs directly in your terminal. Unlike Claude.ai in your browser, it reads and writes files, runs commands, searches your folder structure, and connects to external services. It does not just answer questions. It does things.
For developers, that means writing and testing code. For everyone else, it means having an AI that works in your files and across your tools, not just in a chat window you copy-paste from.
The difference: Claude.ai is a conversation. Claude Code is a colleague who has access to your drives and calendar and goes and builds the thing, not just advises on it.
Which Claude Code terms come up most often?
The glossary below covers 28 terms grouped alphabetically. Each entry answers one question in 40 to 80 words. Read top to bottom, or jump to the letter you need.
What is an agent (subagent) in Claude Code?
An agent, officially a subagent, is a specialised assistant Claude hands one job to. It runs in its own context window, with its own instructions and its own set of tools, and sends a short summary back to your main conversation. Use one to keep a big search out of your main chat, or to run several jobs side by side. Each subagent is a Markdown file in .claude/agents/: a job description, not a prompt.
Why you care: A subagent gives one job its own specialist, without cluttering the chat you are working in.
What does AGENTS.md do?
AGENTS.md is a project instruction file that works across AI coding tools, not just Claude Code. If your project has one and no CLAUDE.md, Claude Code reads it as your instructions. Handy when your team also uses other AI tools: you write the rules once and every tool reads the same file.
Why you care: If colleagues use other AI coding tools, one shared rules file keeps everyone working from the same instructions.
What is a branch in Git?
A branch is a separate version of your files that runs alongside the main one. You make changes there, check that they work, and merge them back in. Nothing breaks until you choose to apply it. Most Claude Code work happens on branches, not directly on the main line.
Why you care: You can try a bold change to your setup without risking the version that already works.
What is Claude Code?
Claude Code is Anthropic's AI assistant that runs inside VS Code or your terminal. It reads files, runs commands, searches your folder structure, and connects to external services through MCP. Different from Claude.ai in your browser: Claude Code does the work in your files, not just in a chat window. The rest of this glossary describes the parts that make it work.
Why you care: This is the step from asking AI for advice to letting it do the work in your own files.
What does CLI mean?
CLI stands for Command-Line Interface. You control a CLI by typing commands instead of clicking menus. Examples: git status, npm install, gh pr create. Claude Code uses the CLI heavily. You do not need to master it yourself, but knowing it exists explains a lot of what you see.
Why you care: When Claude shows you a command, you know it is an instruction to your computer, not an error.
What is CLAUDE.md?
CLAUDE.md is the instruction file Claude Code reads at the start of every conversation. It tells Claude who you are, what the project is, which rules apply, and where to find things. Not a knowledge dump, a map. One of the most useful files you set up.
Why you care: This one file decides whether Claude knows your work or starts from zero every session.
What is a commit?
A commit is a saved snapshot of your work with a message explaining what changed. Like a version in document history, but intentional: you decide when to commit and what to write. "Updated homepage copy" or "Fixed broken link in nav". Small frequent commits beat one big commit at the end.
Why you care: Every commit is a point you can return to when an experiment goes wrong.
What is compaction?
Compaction is what happens when your conversation gets close to the limit of the context window: Claude Code summarises the conversation so far and carries on from that summary. It happens automatically, or you trigger it with /compact. Your CLAUDE.md survives compaction. Instructions you only typed in the chat may not, which is why rules you care about belong in CLAUDE.md.
Why you care: If Claude suddenly forgets a rule from earlier in the chat, compaction is usually why, and CLAUDE.md is the fix.
What is the context window?
The context window is everything Claude holds in mind during a session: your conversation, the files it read, command output, CLAUDE.md and loaded skills. Think of it as working memory, and it is not infinite. When it fills up, Claude Code compacts the conversation into a summary. Type /context to see what is taking up space, or /clear to start fresh.
Why you care: A crowded context window gives vaguer answers, so a fresh session per task often beats one endless chat.
What is a draft in Claude Code output?
A draft is what Claude produces on a first pass. Treat it as work from a smart colleague who has not read everything twice. Check it before you use it. The quality of the output depends on the quality of the input, which is your prompt plus your CLAUDE.md context.
Why you care: You stay the one who signs off, which is what makes it safe to hand AI real work.
What is a VS Code extension?
An extension is a plugin that adds functionality to VS Code. Claude Code runs as an extension inside VS Code, which is why you use it without leaving your editor. You install it once from the VS Code marketplace and it stays connected to your workspace.
Why you care: You can use Claude Code in the editor you already have open, without living in a terminal.
What is Git?
Git is the version control system that tracks changes to your files over time. It keeps full history, lets you go back, and makes it possible for multiple people to work on the same project without overwriting each other. Claude Code uses Git constantly in the background. You do not need to learn all of it, but it explains the terms branch, commit, push and pull.
Why you care: Git is your undo button for a whole project, not just for the last sentence you typed.
What is a hook in Claude Code?
A hook is an instruction that runs automatically when something happens. Example: every time Claude Code starts a session, run git pull to make sure you have the latest version. You set up hooks once and they fire without you thinking about it. Good for routine tasks you would otherwise forget.
Why you care: Hooks turn "remember to do X" into something that happens by itself.
What is INDEX.md?
INDEX.md is a short file that describes what is in a folder. It helps Claude (and you) find things without reading every file. Maintained automatically by the index-builder skill, not something you update by hand. Keep it under 200 lines.
Why you care: A good map means Claude finds the right file faster and reads less of what it does not need.
What does MCP stand for?
MCP stands for Model Context Protocol. It is the standard that lets Claude talk to external services: your calendar, task manager, email, or internal tools. Each service runs as an MCP server, and Claude picks up tools from it. When someone says "I connected Claude to ClickUp", that is MCP under the bonnet.
Why you care: MCP lets Claude work in your calendar, mail and task list, so you stop copy-pasting between them.
What is MEMORY.md?
MEMORY.md is an index of persistent notes Claude carries between conversations. Not the notes themselves, just the list of what exists and where. This means Claude remembers your context, preferences and ongoing projects without you having to re-explain everything each session.
Why you care: You correct Claude once, and it keeps that lesson for next time.
What is permission mode?
Permission mode controls how strict Claude is about asking your approval before doing things. On strict settings, Claude asks before every file write. On looser settings, it works more autonomously. You choose the balance based on how much you trust the setup and what the task involves.
Why you care: You decide how much freedom Claude gets, task by task. Start strict and loosen up as your trust grows. Switch with Shift+Tab.
What is plan mode?
Plan mode is a setting where Claude shows you what it intends to do before doing it. Useful for anything non-trivial. Claude presents the plan, you approve or adjust, then it executes. Catches misunderstandings before they become file changes.
Why you care: Reading a plan first is faster than undoing a wrong turn afterwards.
What is a good prompt for Claude Code?
A good prompt is short and contextual. "Write a summary of this meeting in three bullet points for a non-technical audience" beats a paragraph about why you need it and what you have tried before. Claude already knows a lot from CLAUDE.md. Use your prompt for the specific job at hand.
Why you care: Once your context lives in CLAUDE.md, your prompts get shorter and your results get better.
What is a pull request (PR)?
A pull request is a formal proposal to merge changes from one branch into another, usually reviewed before it goes through. Common in teams working on shared files. If you ever see "can you review my PR", this is what they mean. Created with gh pr create from Claude Code.
Why you care: A PR lets a colleague check the work before it goes live, the way you would review a draft.
What is a remote in Git?
A remote is the copy of your files stored outside your computer, usually on GitHub. Your local files are what you work in. The remote is the backup and the shared source. push sends local changes to the remote. pull brings remote changes to your local copy.
Why you care: A remote means your work survives a lost laptop and can be picked up on another device.
What is a repository (repo)?
A repository is a project folder with full version history attached. Everything that has ever been committed is still there, traceable. Your repo is what Claude Code works in. One workspace, one repo, usually.
Why you care: Everything Claude changes is traceable, so you can always see what happened and when.
What is a shell?
A shell is the programme that interprets the commands you type in a terminal. On Windows that is PowerShell. On Mac and Linux that is Bash. The terminal is the window. The shell is what understands what you type. Different shells use different syntax, which is why some commands look different across systems.
Why you care: When a command works on a colleague's Mac but not on your Windows laptop, the shell is usually the reason.
What is a skill in Claude Code?
A skill is a reusable instruction set, saved as a SKILL.md file. Each skill does a specific job: writing a LinkedIn post, summarising a document, checking your calendar. You call one by typing /name, and Claude also picks one up by itself when it fits the task. Think of skills as saved recipes for work you do more than once.
Why you care: A skill turns a task you explain every week into one you never have to explain again.
What is a slash command?
A slash command, now officially just a command, is anything you type starting with /. Built-in ones control the session: /clear, /compact, /model. Your own skills show up there too, such as /news or /reflect. Type / in Claude Code and you see the full list. To build your own, make a skill: that is the recommended way.
Why you care: Type / and you see everything your setup can do. It is the quickest way to learn it.
What is a system prompt?
A system prompt is the fixed set of instructions Claude Code sends ahead of your conversation every time: how to use its tools, how to behave safely, how to format answers. You usually do not see it, but it shapes everything. Your CLAUDE.md is not part of it. Claude Code adds CLAUDE.md to the conversation as a separate message, right after the system prompt.
Why you care: It explains why Claude Code behaves differently from the chat on claude.ai, even with the same model.
What is the terminal?
The terminal is the window where you type commands. In VS Code you open it with Ctrl+`. The terminal itself is just the window. The shell is what runs inside it and interprets what you type. On a Windows workspace the default shell is PowerShell.
Why you care: You will see it a lot, but you rarely need to type in it yourself.
What is a workspace in VS Code?
A workspace is the root folder VS Code and Claude Code consider your working environment. Everything within it is accessible to Claude. Everything outside it is not, unless you explicitly give access. Your workspace is where CLAUDE.md lives, which means it is where Claude's instructions live.
Why you care: Pick your workspace folder with care: that is exactly what Claude can see and change.
Which Claude Code terms have been renamed?
Claude Code moves fast, and some names in older tutorials, videos and blog posts are out of date. If a guide uses a word you cannot find in the app, check this list first.
| Old name | Now called | What changed |
|---|---|---|
| Slash commands | Commands | Same thing. Anthropic dropped the word "slash". |
| Custom commands | Skills | Old files in .claude/commands/ still work, but skills are the recommended way to build your own. |
| Headless mode | Non-interactive mode | Same -p flag, same behaviour: run one prompt and stop, for scripts and automation. |
| Web session | Cloud session | A session that keeps running on Anthropic's servers after you close your laptop. "Claude Code on the web" now only means the browser version. |
What does this actually mean for you?
You do not need to understand all of this to start using Claude Code. You need to understand enough to have a conversation with someone who does.
The terms above come up constantly. Knowing what a hook is, why the context window matters, or what the difference is between a skill and an agent, saves you from nodding along while someone explains their setup.
More practically: Claude Code is not a developer tool that happens to be useful for other people. It is a system for giving AI access to your actual work, with guardrails you define. That is useful whether you write code or not.
The people getting the most out of it are not the most technical. They are the ones who set up CLAUDE.md properly, think about what context Claude actually needs, and decide which tasks are worth automating and which are not.
When you are ready to move beyond a glossary into a real working setup, the BUILD framework gives you the five steps to create your first AI team member: pick the task, write the rules, gather the knowledge, lay out the instructions, debug until it works. The frame applies whether you use Claude Code or a Custom GPT.
Frequently asked questions
Do I need to know how to code to use Claude Code?
No. Claude Code is built for technical work, but most of the value for a non-developer comes from setting up CLAUDE.md, configuring agents and skills, and connecting MCP servers. None of that requires writing code yourself. You write instructions in plain English (or Dutch), and Claude executes them.
How is Claude Code different from Claude.ai?
Claude.ai runs in your browser and works inside the chat: it answers, writes and can use the apps you connect, but it does not touch the files on your computer. Claude Code runs inside VS Code or your terminal and reads files, writes files, runs commands, and connects to external services through MCP. The difference is access. Claude.ai is a conversation. Claude Code does the work in your files.
What is the first file I should set up in Claude Code?
CLAUDE.md, in the root of your workspace. It tells Claude who you are, what the project is, what rules apply, and where to find things. A good CLAUDE.md is a map, not a knowledge dump. Without it, Claude works in the dark. With it, every session starts with shared context.
Can Claude Code connect to my calendar, email, or task manager?
Yes, through MCP servers. Common integrations include Google Calendar, Gmail, ClickUp, Notion, and GitHub. Each integration runs as a separate MCP server you configure once. After that, Claude picks up tools from those servers and uses them in your conversations. You control which services are connected.
Is Claude Code safe to use on confidential work?
It depends on your setup. Claude Code runs locally and only sends what you tell it to send. Permission mode controls what gets approved before execution. MCP servers run with your credentials. Confidential work is fine on a properly configured setup. Read the security guidance before you connect production systems.
This glossary was assembled by Guus Witjes at Step Ahead AI, based on running Claude Code as a daily tool since November 2025, across HAN University's Digital Marketing team and his own consulting work.