# Claude Code Glossary

> Everything you need to know to get started with the most popular AI coding tool on the market

Canonical URL: https://antoniodevivo.com/en/blog/2026-10-22-claude-code-glossary/
Language: en
Author: Antonio De Vivo
Published: 2026-10-22T00:00:00.000Z
Updated: 2026-10-22T00:00:00.000Z
Tags: claude, code, claude-code, ai, gen-ai, glossary, coding, vibe, vibe-coding, key-concepts, concepts
Version: 1
Translations: [it](https://antoniodevivo.com/it/blog/2026-10-22-claude-code-glossary/)

Getting into the world of Claude Code and its ecosystem seems easy:

-   open a terminal
-   type `claude`
-   start firing off prompts like crazy

If you’re using Claude Code (which I’ll call CC from now on) this way, you’re only tapping into 20% of its potential.

In fact, Anthropic provides a whole range of basic tools that let you configure CC and make it more powerful.

And because it’s easy to get lost among all these terms, I thought I’d put together a short glossary to help you get a feel for the CC ecosystem!

Ready? Great, let’s get started!

## Features and extensions

| Name | Definition | How to activate it | How it works |
| --- | --- | --- | --- |
| Agent | The agent that manages the conversation with the user and coordinates work on the project. | It comes into play when you submit a request in a Claude Code session. | It interprets the goal, gathers information, uses tools, checks results, and can delegate tasks to subagents. |
| Subagent | A secondary agent you can delegate a specific task to, with its own context separate from the main agent’s and dedicated instructions. It reports its results back to the agent that launched it. | The main agent launches it when delegation would be useful, or when you explicitly ask for it in your prompt. Custom subagents can be defined in Markdown files under `.claude/agents/`. | It uses the tools permitted within its own context and returns the result to the main agent. It can work while the main agent waits. |
| Skill | A Markdown file containing reusable knowledge and instructions that Claude loads and uses when needed. | The user can invoke it with `/skill-name`, or Claude can activate it when needed, depending on the configuration. | It is loaded into the context and provides useful information and solutions. |
| Hook | A file used to configure an automation that runs when a Claude Code event occurs. | It is triggered by a CC event, such as before using a tool or after making a change. | It can launch scripts or other configured actions. |
| Script | A file written in a programming language, usually JavaScript or TypeScript.  <br>It performs specific actions without depending on Claude’s interpretation. | By convention, it is defined in the `.claude/scripts` folder and invoked by skills, hooks, or agents. | Claude executes it through the terminal. |
| Tool | An operational capability Claude can use, such as reading or editing a file, searching through code, or running a terminal command. | When needed. | Generally through the terminal or as a Claude Code tool. |
| MCP (Server) | A suite of tools written in programming languages that enable interaction with external platforms or simply execute application logic. | CC uses them as needed and in response to explicit user requests. | Claude queries the server, which provides a list of tools with names and descriptions, along with documentation for the entire suite.  <br>Claude reviews this information and decides which tools to call and in what order. |
| Plugin | An installable package that bundles one or more extensions, such as skills, subagents, hooks, and MCP servers. It makes it easier to distribute and reuse a suite of configurations and tools. | For example, install it with `/plugin install plugin-name@marketplace-name`. | The suite works together with CC. |
| Marketplace | A catalog where you can discover and install plugins. It is the distribution channel, not the plugin itself. | In the CC chat, type: `/plugin marketplace add marketplace-name` | Claude Code reads its catalog and makes the plugins available for installation. Adding a marketplace does not automatically install all the plugins it contains. |
| Slash command | A command that starts with `/`. It can control the session, like `/clear`, or invoke an available skill. | In the CC chat, type `/`. | It depends on the command. |

* * *

## Main files and folders

| Name | Type | Definition | How to activate it | How it works |
| --- | --- | --- | --- | --- |
| CLAUDE.md | File | A Markdown file containing persistent instructions and information: conventions, project structure, and useful commands. | Place it in your project’s root directory. | It enters the context window at the start of every session. |
| SKILL.md | File | The main file of a skill: it contains metadata, a description, and instructions. It can be accompanied by scripts, examples, and other materials. | Create a folder named after the skill under `.claude/skills`.  <br>Example: `.claude/skills/skill-name/SKILL.md` | Claude automatically uses it when needed. |
| .claude | Folder | The folder that holds Claude Code configurations and extensions. | Create it in your project’s root directory. | CC automatically recognizes and uses it. |
| ~/.claude/settings.json | File | The user’s global settings, applied across different projects on the same machine. | Create it in the `.claude` folder under your user directory.  <br>Example: `C:/Users/Alfredo/.claude/settings.json` | It is automatically used whenever CC reloads. |
| .claude/settings.json | File | Project settings, which can be version-controlled and shared with the team. | Create it in your project’s `.claude` folder.  <br>Example: `C:/prj-1/.claude/settings.json` | It is automatically used whenever CC reloads. In case of conflicts with the global file, it takes precedence according to each setting’s rules. |
| .claude/settings.local.json | File | Personal settings that apply only to that project and are intended to remain outside version control. | Create it in your project’s `.claude` folder.  <br>Example: `C:/prj-1/.claude/settings.local.json` | It is automatically used whenever CC reloads. In case of conflicts with the global file, it takes precedence according to each setting’s rules. |
| .claude/skills/ | Folder | The project’s skills folder: each skill has its own subfolder containing its `SKILL.md` file. | Create it in your project’s root directory. | CC uses it when needed. |
| .claude/agents/ | Folder | The folder containing definitions for custom project subagents. | Create it in your project’s root directory. | CC uses it when needed. |
| .claude/commands/ | Folder | The legacy location for defining custom commands in Markdown, which is still supported. Skills now provide a more comprehensive mechanism for these workflows. | Create it in your project’s root directory. | The file’s contents are used as instructions for Claude when you invoke the command. They are not executed directly as code: Claude interprets the instructions and uses the necessary tools. |
| ~/.claude.json | File | A file managed by Claude Code to store internal state and configurations, such as login information and certain MCP configurations. It is separate from `settings.json`. | Claude Code normally creates and manages it during use. | It is updated over time with the user’s new global configurations.  <br>It does not contain absolutely all global settings: many are stored in `~/.claude/settings.json`. |

* * *

## Context and memory

| Name | Definition | Hints & Tips |
| --- | --- | --- |
| Session | A working conversation with Claude Code, with its own history and context. It can be resumed later. | With the terminal command `claude --resume`, you can find the history of sessions held in a given folder for the current user. |
| Context | The information available to the model at a given moment: instructions, messages, content it has read, and tool results. | N/A |
| Context window | The maximum amount of information the model can process at once, measured in tokens. | It is the sum of input tokens (prompts, skills, and other configuration text) and output tokens.  <br>  <br>The fuller a context window gets, the harder it becomes for the LLM to maintain good output quality. |
| Token | A unit the model uses to represent and measure text. Generally, a token is part of a word, somewhat like splitting a word into syllables. Tokens are used to measure context, usage, and, where applicable, API costs. | Output tokens usually cost at least three times as much as input tokens. |
| Compaction | Summarizing the conversation to free up context space while retaining relevant information. You can request it with `/compact`. | A best practice is not to rely too heavily on this method of trimming the context window.  <br>Instead, it is preferable to:  <br>• have Claude save relevant information in one or more Markdown files  <br>• use `/clear` to reset the context window  <br>• tell Claude to use the information in those Markdown files as the basis for continuing the work |

* * *

That’s it!

I hope this has helped clear things up a little. I’ve tried to make the concepts as straightforward and concise as possible.

If you’re still finding things difficult, try putting these ideas into practice as you go, working with the Claude Code interface and focusing on one concept at a time.

I’m sure you’ll soon realize it wasn’t so difficult after all.

Sending you a hug, and happy coding!
