Glossary

Last verified: 13 February 2026 | Applies to: All plans

In 30 seconds

Claude’s platform has its own vocabulary. This glossary defines the terms you’ll encounter throughout the Operator’s Toolkit and in Claude itself. Sorted alphabetically.

Artifact

An interactive output that Claude creates and renders directly in the chat window — a chart, a document, a code snippet, a table. Artifacts can be downloaded, copied, or iterated on. They’re how Claude shows you work rather than just describing it.

CLAUDE.md

A markdown file you place in a folder to give Cowork persistent instructions. When Claude works in that folder, it reads this file automatically. Used for folder-specific context like team info, project rules, and terminology. Also created by the Productivity plugin’s /start command.

Claude Code

The CLI (command-line interface) version of Claude for developers. Installed via npm or Homebrew. Lives in your terminal, understands codebases, and helps with development through natural language. Not the same as the Code tab (which is the visual version).

Claude in Chrome

A Chrome browser extension (Chrome only — not Edge, Brave, or other browsers) that lets Claude see, interact with, and automate web pages. Features include workflow recording, scheduled tasks, and multi-tab management. Pro users are limited to Haiku 4.5.

Code tab

The visual version of Claude Code inside Claude Desktop. Same capabilities as the CLI but with a graphical interface. Good for operators who want coding power without opening a terminal.

Connector

An integration that links Claude to an external tool or service via MCP. Each connector authenticates via OAuth and gives Claude scoped access to read from or write to that tool (e.g., Slack, Asana, Notion). Browse available connectors at claude.com/connectors.

Context window

The amount of text Claude can process in a single conversation, measured in tokens. Currently 200K tokens (roughly 150,000 words) for all models. Opus 4.6 has a 1M token context available in beta with tier 4 API access.

Cowork

Claude’s autonomous mode. Runs in Claude Desktop on macOS and Windows x64 (arm64 not supported). Gives Claude the ability to read files, browse the web, execute code, and take multi-step actions — all in a sandboxed environment on your computer. Requires Pro or above.

Extended thinking

A Chat feature where Claude shows its reasoning process before answering. Useful for complex analysis where you want to see how Claude arrived at its conclusion, not just the final answer.

Global instructions

Persistent instructions that apply to every Cowork session, regardless of which folder you’re working in. Set via a CLAUDE.md file at a global level.

Haiku 4.5

Claude’s fastest, most lightweight model. Best for simple tasks, quick lookups, and high-volume processing. The only model available in Claude in Chrome for Pro plan users.

MCP (Model Context Protocol)

An open standard that defines how Claude communicates with external tools and services. MCP is the protocol; connectors are the specific integrations built on it. Maintained at modelcontextprotocol.io.

MCP App

An extension to MCP that lets connectors render interactive UI elements directly in Claude’s chat window — buttons, forms, cards, charts. Current partners include Slack, Asana, Figma, Canva, Box, Hex, monday.com, Notion, Amplitude, and Clay.

Opus 4.6

Claude’s most capable model. Best for complex reasoning, nuanced analysis, and long-form content. Slowest and highest-cost per message. The 1M token context is beta-only (requires tier 4 API access).

Plugin

A pre-built bundle that packages skills, connectors, slash commands, and sub-agents for a specific business function. Anthropic released 11 knowledge-work plugins (Productivity, Finance, Legal, Sales, Marketing, Data, Product Management, Customer Support, Enterprise Search, Bio Research, and Plugin Create). All open source. Install from claude.com/plugins or the Cowork interface.

Plugin Create

One of the 11 knowledge-work plugins. Used to build new plugins from scratch — describe what you want, and Claude generates the full plugin structure. Also includes Plugin Customiser for adapting existing plugins.

Progressive disclosure

How Claude loads skills — it reviews what’s available, determines what’s relevant to your request, and loads only the necessary instructions. This keeps responses focused without overloading context.

Project

A feature in Chat that groups related conversations and files together. Projects let you set system instructions and upload reference documents that apply to all conversations within the project.

Sandbox

The isolated environment on your computer where Cowork runs. Claude can access the folder you selected and its subfolders, but not the rest of your system.

Skill

A folder containing markdown instructions (with YAML frontmatter), optional executable scripts, and supplementary resources that give Claude domain expertise. Skills are the knowledge layer — they can exist inside plugins or independently. Available on all plans including Free.

SKILL.md

The required file inside a skill folder. Contains YAML frontmatter (name, description) and markdown instructions that Claude uses for domain-specific tasks.

Slash command

A shortcut that triggers a specific action in a plugin. Typed as /command-name in a Cowork session (e.g., /start, /reconciliation, /review-contract).

Sonnet 4.5

Claude’s balanced model — good performance at moderate speed and cost. The recommended default for most operator tasks.

Sub-agent

A specialised agent within a plugin that handles specific multi-step tasks. Sub-agents run autonomously within the scope of their plugin.

Token

The unit Claude uses to measure text. Roughly 0.75 words per token. A 200K token context window is approximately 150,000 words.

What Is Claude? — platform overview
Plugins Overview — all 11 plugins explained
Connectors — MCP integrations
Skills — the knowledge layer

Something wrong or outdated? Let us know →

Get weekly workflows — subscribe to the newsletter.