Reference
The AI Tools Glossary
Short, accurate definitions of the terms you keep running into across Claude Code, OpenAI, and Cursor: Skills, hooks, MCP, the OpenAI API, GPT, Cursor Tab, and more.
Agent Skills
Agent Skills are reusable, file-based instruction packages that Claude Code loads on demand to give the agent specialized, repeatable expertise for a task.
AI IDE
A code editor built around AI assistance (autocomplete, chat, and agents), with Cursor being the best-known example.
Background Tasks
Background Tasks let Claude Code run long-lived commands like dev servers and watchers without blocking the chat, then read their output later.
ChatGPT
OpenAI's conversational AI assistant, available as a web, desktop, and mobile app, widely used by developers for explaining code, drafting, and pair-programming.
Checkpointing
Checkpointing is a Claude Code feature that automatically snapshots your code before each edit, so you can rewind your files and conversation to an earlier point.
CLAUDE.md
CLAUDE.md is a Markdown file Claude Code reads at startup to load project-specific context, conventions, and instructions into every session.
Context Compaction
Compaction is how Claude Code condenses an older, growing conversation into a summary so a long session can keep going without exceeding the context window.
Context Window
The context window is the finite span of tokens — your prompts, files, tool output, and Claude's replies — the model can consider at once in a Claude Code session.
Cursor Composer
Cursor's multi-file, agentic editing surface for making larger changes across a codebase from a natural-language prompt.
Cursor Tab
Cursor's predictive multi-line autocomplete that suggests edits as you type and lets you accept them with the Tab key.
Extended Thinking
Extended Thinking is a Claude Code mode where Claude reasons through a problem step by step before answering, improving accuracy on hard tasks.
GPT
OpenAI's family of large language models (the engines behind ChatGPT and much of the OpenAI API), updated and versioned over time.
Headless Mode
Headless mode runs Claude Code non-interactively from the command line, taking a prompt as input and printing the result so it can be scripted and automated.
Hooks
Hooks are user-defined shell commands that Claude Code runs automatically at specific lifecycle events, such as before a tool runs or when a session ends.
MCP (Model Context Protocol)
MCP (Model Context Protocol) is an open standard that lets Claude Code connect to external tools and data sources through pluggable MCP servers.
Memory
Memory is how Claude Code persists instructions across sessions, loaded automatically from CLAUDE.md files at the project, user, and enterprise scopes.
OpenAI API
OpenAI's developer platform for building with its models, with its own changelog covering new endpoints, models, parameters, and deprecations.
Output Styles
Output Styles modify Claude Code's system prompt to change its role, tone, and response format — for example switching it into an explanatory or learning mode.
Permissions
Permissions are Claude Code's rules that control which tools and actions Claude runs automatically versus what pauses to ask for your approval.
Plan Mode
Plan Mode is a read-only Claude Code mode where Claude researches your codebase and proposes a plan first, without editing files, until you approve it.
settings.json
settings.json is the JSON config file where Claude Code stores settings like permissions, environment variables, and hooks at the user, project, or local level.
Slash Commands
Slash commands are shortcuts in Claude Code, starting with "/", that trigger built-in actions, bundled skills, or reusable custom prompts you define.
Subagents
Subagents are specialized Claude Code assistants, each with its own context window, system prompt, and tools, that handle focused tasks on the main agent's behalf.