.claude/rules/
A folder with rules that load only when Claude works on certain files. Like a note stuck only on the drawer it’s about.
From module 02 · CLAUDE.md: your project’s memory →Terms, explained simply
C
E
.env
A hidden file in the project where passwords and secret keys are kept. It never goes into git, and it’s a good idea to forbid Claude from reading it.
From module 03 · Permissions: what Claude may do →G
.gitignore
A list of files that git must ignore. What you write there doesn’t reach the history or your colleagues, for example your personal settings.
From module 02 · CLAUDE.md: your project’s memory →M
.mcp.json
The file in the project where you write which MCP servers (connections to other apps) the team uses. Passwords don’t go in it; they’re taken from environment variables.
From module 08 · MCP: connecting to other apps →C
/clear
The command that empties the current conversation and starts from zero. Use it when you move on to a new task: Claude works better and it costs less.
From module 01 · How to talk to Claude →/compact
The command that summarizes the long conversation so far, to free up room. You can say what to keep, for example “/compact keep the decisions about the API”.
From module 04 · The working method, step by step →I
/init
The command that has Claude look through the project and write a first CLAUDE.md file on its own, which you then shorten and correct.
From module 02 · CLAUDE.md: your project’s memory →A
acceptEdits
A working mode in which Claude changes files without asking you anymore, but still asks permission before running commands.
From module 03 · Permissions: what Claude may do →Agent
An artificial intelligence program that doesn’t just answer, but also takes steps by itself: it reads files, runs commands, makes changes. You give it the task, and it sees it through.
From module 00 · Installation and first steps →Agent SDK
A library (ready-written code) with which programmers put Claude Code’s “engine” into their own applications.
From module 10 · Claude on autopilot →Allowlist
The list of actions that Claude can take without asking you, for example running the tests. Like a list of guests who come in without ringing the bell.
From module 03 · Permissions: what Claude may do →API
The way two programs talk to each other. It’s like a restaurant’s menu: it shows what you can ask for and what you get back.
API key
A long password that gives a program access to a service, for example to Claude. You pay for what you use. It’s kept secret, like a PIN.
From module 00 · Installation and first steps →B
Bash
One of the “languages” in which you write commands in the terminal. When you see “Bash(...)” in the settings, it’s about the commands that Claude can run.
Branch
A separate version of the project in git, where you work on something new without breaking the main version. It’s not a copy in another folder: in one folder you’re on a single branch at a time. When you’re happy, you merge the branch into the main version.
From module 04 · The working method, step by step →Build
The step in which the code written by a person is “assembled” into a form ready to run. If the build fails, something in the code doesn’t fit together.
bypassPermissions
The mode in which Claude does anything without asking anymore. Dangerous: it’s used only in an isolated container, with no important data.
From module 03 · Permissions: what Claude may do →C
Checkpoint
A point saved automatically before Claude changes files. You can go back to it, like “Undo”, but it doesn’t replace git.
From module 04 · The working method, step by step →CI
A robot that automatically runs checks (tests, review) every time someone sends new code to GitHub.
From module 10 · Claude on autopilot →Claude Code
The Anthropic assistant that works directly in your project: it reads files, runs commands and writes code, asking your permission first.
From module 00 · Installation and first steps →CLAUDE.local.md
The personal version of CLAUDE.md, only for your computer. It isn’t shared with colleagues. For example: “answer me in Spanish”.
From module 02 · CLAUDE.md: your project’s memory →CLAUDE.md
A text file that Claude reads at the start of every session. It’s like an instruction sheet stuck on the fridge for a new colleague.
From module 02 · CLAUDE.md: your project’s memory →CLI
A program that you use by typing commands in the terminal, not by clicking. “claude” and “gh” are CLIs.
Commit
A save of the project in git, with a message that says what changed. Like a save in a game, which you can go back to at any time.
From module 04 · The working method, step by step →Container
A “computer in a box”, isolated from the rest of the system. What breaks there doesn’t touch your real files.
From module 03 · Permissions: what Claude may do →Context
Everything Claude “remembers” in the current conversation: messages, files read, results. It has limited room, like a workbench: too cluttered, and it’s hard to work.
From module 04 · The working method, step by step →Conventional Commits
A rule for writing commit messages, with a word at the start: “feat:” for something new, “fix:” for a repair, “docs:” for documentation.
From module 04 · The working method, step by step →D
Database
An organized place where an application keeps its data, like a cabinet with labeled drawers: users, orders, products.
From module 08 · MCP: connecting to other apps →Debugging
Searching for and fixing a mistake in the program: you reproduce the problem, find the cause, correct it and check.
Deny
The list of actions that Claude is never allowed to take, for example reading the passwords. The prohibition always wins.
From module 03 · Permissions: what Claude may do →Dependency
Code written by others that your project uses, like the parts bought from a store for a piece of furniture.
Deploy
Publishing the project so others can use it, for example putting a website on the internet.
Dev server
A small server started on your computer, so you can see the website in the browser while you work on it (usually at localhost).
Diff
The exact list of changes: which lines were deleted and which lines were added. Read it before you accept the changes.
From module 04 · The working method, step by step →E
Edge cases
The unusual situations in which a program can go wrong: an empty list, the number zero, a very long text. Good tests check them.
Endpoint
An “address” of a server to which a program can send a request, for example POST /users to create a user.
Environment variable
A named value that programs read when they start. You can set it in the system (e.g. from the terminal) or in Claude’s settings (the “env” section). Secret keys you keep only on your own computer, never in files that reach git.
From module 03 · Permissions: what Claude may do →Exit code
A number that a script returns at the end to say how it went: 0 means “everything is fine”, 2 means (in hooks) “stop the action”.
From module 07 · Hooks: automatic actions →F
Flag
An option added to a command, which starts with “-” or “--”. For example, in “claude --model opus”, “--model” is a flag.
From module 10 · Claude on autopilot →Folder
A place on the computer where files and other folders are kept. A project is, in fact, a folder.
Formatter
A program that tidies the code automatically (spaces, lines, quotation marks), so it looks neat and the same everywhere.
From module 07 · Hooks: automatic actions →Frontmatter
A few lines of settings at the start of a file, between two “---” lines. They give, for example, a skill’s name and description.
From module 05 · Skills: reusable recipes →G
gh CLI
GitHub’s official program for the terminal. With it, Claude can read issues and open pull requests.
From module 04 · The working method, step by step →Git
A program that keeps the project’s history: every save (commit) stays and you can go back to it. Like the saves in a game.
From module 04 · The working method, step by step →GitHub
A website where you keep your git projects online, work with others and discuss changes. Like a Google Drive for code.
GitHub Actions
The robots on GitHub that start by themselves on certain events, for example a new PR, and run the steps you wrote, Claude included.
From module 10 · Claude on autopilot →Glob
A pattern for naming several files at once. “*” means “anything”, and “src/**” means “everything in src, however deep”.
From module 03 · Permissions: what Claude may do →H
Haiku
The fastest and cheapest Claude model. Good for simple tasks, such as searching through files.
From module 03 · Permissions: what Claude may do →Headless
Running Claude without the chat window: you give it the task in a single command (“claude -p”), it does it, gives you the result and stops. Good for automation.
From module 10 · Claude on autopilot →Hook
An automatic action that always starts at a particular moment, for example after every edit. Like an alarm that’s guaranteed to ring, without anyone having to remember.
From module 07 · Hooks: automatic actions →I
IDE
A program for writing code, with lots of helpers, for example VS Code. It’s like Word, but for code.
From module 00 · Installation and first steps →Issue
A ticket on GitHub in which someone describes a problem or an idea. Each one has a number, for example #142.
From module 08 · MCP: connecting to other apps →J
jq
A small terminal program that pulls a specific piece of information out of a JSON text. The starter kit’s hooks need it.
From module 07 · Hooks: automatic actions →JSON
A text format for data and settings, with curly braces and “name”: value pairs. Claude’s settings files are in JSON.
From module 03 · Permissions: what Claude may do →L
Lint
An automatic check that looks for small mistakes and untidy style in the code, like a spell checker for programs.
Localhost
Your computer’s address. A website opened at localhost runs only on your machine, not on the internet.
From module 08 · MCP: connecting to other apps →Lockfile
An automatically generated file that notes the exact versions of the parts the project uses. It isn’t edited by hand.
From module 07 · Hooks: automatic actions →M
Markdown
A simple way of writing formatted text: “#” makes a heading, “-” makes a list. The .md files, like CLAUDE.md, are Markdown.
Marketplace
A catalog of plugins, kept in a git repository, from which the team installs packages with one command.
From module 09 · Plugins: ready-made packages →MCP
A kind of universal adapter through which Claude connects to other apps: GitHub, a browser, a database, Notion.
From module 08 · MCP: connecting to other apps →Migration
A file that changes the structure of the database, for example adds a column. Once applied, it isn’t modified anymore.
N
Node.js
A program that runs JavaScript code on your computer, not just in the browser. Many tools, npm included, need it.
From module 00 · Installation and first steps →npm
A “store” of ready-made code pieces for JavaScript, used from the terminal. npx starts such a piece without installing it; pnpm is an alternative to npm.
From module 00 · Installation and first steps →O
OAuth
The way you sign in to a service by clicking “Allow” in a window, without giving your password to another program.
From module 08 · MCP: connecting to other apps →Opus
The most powerful Claude model, but also the most expensive. Good for planning, architecture and hard problems.
From module 03 · Permissions: what Claude may do →Output
What a program prints or returns after it runs: a text, a result, an error message.
Output style
A setting that changes the way Claude answers you. For example, in the Learning style it explains what it’s doing and lets you write small pieces of code yourself.
From module 11 · Pro tips →P
Pattern
A template or a usual way of doing something. An anti-pattern is a bad habit, which looks good but spoils the result.
Permissions
The rules that say what Claude may do on its own, what it must ask you about and what it’s not allowed to do. Like the keys you give a handyman.
From module 03 · Permissions: what Claude may do →Pipe
The “|” sign in the terminal: it takes what one command wrote and passes it on to another command, like a pipe between them.
From module 01 · How to talk to Claude →Pipeline
A chain of automatic steps that run one after another, for example: check the code, run the tests, publish.
From module 10 · Claude on autopilot →Plan mode
A mode in which Claude only reads and proposes a plan, without changing anything. You correct it before the work starts.
From module 04 · The working method, step by step →Playwright
A tool that controls an automated browser. Through MCP, Claude can open pages, click and take screenshots.
From module 08 · MCP: connecting to other apps →Plugin
A package that gathers skills, subagents, hooks and MCP connections, so you can install them with a single command. Like a ready-made toolbox.
From module 09 · Plugins: ready-made packages →Prompt
The message in which you tell Claude what to do. The clearer it is (what, where, limits, how you check), the better the result.
From module 01 · How to talk to Claude →Prompt injection
A trick in which someone hides instructions in a page or a message, hoping that Claude will obey them. That’s why you don’t give broad rights.
From module 08 · MCP: connecting to other apps →Pull request
A request to add your changes to the project’s main version, so that others check them first. In short: PR.
From module 04 · The working method, step by step →R
Read-only
Access for reading only, with no permission to change anything. Like a display case: you see, but you don’t touch.
From module 08 · MCP: connecting to other apps →Refactoring
Rewriting the code so it’s clearer and tidier, without changing what the program does.
Release notes
The list of what changed in a new version of the program, written for the people who use it.
From module 05 · Skills: reusable recipes →Repository
The folder of a project tracked by git, with its whole history. In short: repo.
From module 04 · The working method, step by step →Review
Someone else (a person or Claude) checking the code before it’s accepted, like a colleague proofreading a text.
From module 06 · Subagents: separate helpers →S
Sandbox
An isolated space in which a program can do what it likes without touching the rest of the computer, like a sandbox for children.
From module 03 · Permissions: what Claude may do →Script
A file with a list of commands that the computer runs automatically, one after another, without you typing them by hand.
Secret
Any piece of information that must not be seen by others: passwords, API keys, tokens. It never goes into git.
From module 03 · Permissions: what Claude may do →Server
A program (or computer) that waits for requests and answers them, for example a website or an MCP connection.
From module 08 · MCP: connecting to other apps →Session
A conversation with Claude, with everything that was discussed in it. It stays saved even after you close Claude, so you can resume it later (for example with “claude -c”). A new session starts from zero.
From module 04 · The working method, step by step →settings.json
Claude Code’s settings file: permissions, hooks, model. The “.local” version is only for you and doesn’t reach git.
From module 03 · Permissions: what Claude may do →Shell
The program inside the terminal that understands and carries out the commands you type.
Skill
A written recipe that Claude follows for a certain type of task. In practice, it’s a folder with a SKILL.md file in it. You start it with /name, or Claude picks it on its own.
From module 05 · Skills: reusable recipes →Slash command
A command that starts with “/”, typed in the conversation with Claude, for example /clear or /init.
From module 05 · Skills: reusable recipes →Sonnet
The “everyday” Claude model: a good balance of speed, cost and quality.
From module 03 · Permissions: what Claude may do →Stack
The set of technologies used in a project, for example “TypeScript + Node.js + PostgreSQL”.
Status line
A line of information shown at the bottom of Claude Code: the model, the branch, the session’s cost.
From module 11 · Pro tips →stderr
The channel on which a program writes its error messages. In hooks, what you write here reaches Claude as an explanation.
From module 07 · Hooks: automatic actions →stdin
The channel through which a program receives data from another. Hooks receive on stdin the details of Claude’s action.
From module 07 · Hooks: automatic actions →stdout
The channel on which a program writes its ordinary result. When you run a command, it’s what you see in the terminal. In hooks, what the hook writes here is read by Claude Code (at session start, for example, it reaches Claude).
From module 07 · Hooks: automatic actions →Subagent
A separate helper that works for Claude, with its own working memory. It gets a task, does it off to the side and returns only the conclusion, like a colleague sent to the archive.
From module 06 · Subagents: separate helpers →System prompt
The basic, hidden instructions that tell an agent who it is and how to behave, before any message from you.
From module 06 · Subagents: separate helpers →T
TDD
A way of working: first you write the tests that describe what the program must do, then the code that makes them pass.
From module 04 · The working method, step by step →Terminal
The window where you type commands as text, instead of clicking. On a Mac it’s called Terminal, on Windows you use PowerShell.
From module 00 · Installation and first steps →Test
A small program that automatically checks that a part of the code does what it should. The test “fails” if it finds a mistake.
From module 04 · The working method, step by step →Token
A small piece of text (roughly a short word) used to measure how much Claude reads and writes, and so the cost. Careful: “token” also means an access password.
Tool
Something Claude can use to act: reading a file (Read), editing it (Edit), running a command (Bash).
Typecheck
An automatic check that looks for the places where the code mixes things that don’t fit, for example a text where a number was needed.
U
UI
The part that a person sees and uses: screens, buttons, forms. It stands for “user interface”.
W
Workflow
Your way of working, step by step. On GitHub, a workflow is also a .yml file that says what the robots should do automatically.
From module 04 · The working method, step by step →Worktree
A second working copy of the same project, in another folder, so two Claude sessions can work in parallel without getting in each other’s way.
From module 11 · Pro tips →Y
YAML
A text format for settings, based on lines and spaces at the start. GitHub Actions files are written in YAML.
From module 10 · Claude on autopilot →