Claude Code Glossary

Essential Claude Code terms grouped by basics, commands, features, settings, and troubleshooting.

• /clear

  Commands

  Clears the current session's history so you can start over from a clean state. Useful when the context has become noisy after a long conversation.

• /compact

  Commands

  Summarizes and compacts the conversation history. Calling it manually reduces token usage while keeping the important details.

• /config

  Commands

  Interactively edits Claude Code settings such as model, theme and permissions without having to touch settings.json by hand.

• /cost

  Commands

  Shows how many tokens the current session has used and the estimated cost so far. Handy when you need to keep spend in check.

• /doctor

  Commands

  A diagnostic command that checks your installation, dependencies and environment. Run it when something feels off to narrow down the problem.

• /exit

  Commands

  Ends the current Claude Code session. An explicit alternative to hitting Ctrl+C twice.

• /help

  Commands

  Shows the list of available slash commands with short descriptions. A good first stop when you forget how something works.

• /init

  Commands

  Analyzes the current project and scaffolds a CLAUDE.md file for you. A convenient first step when onboarding Claude Code to a new repository.

• /memory

  Commands

  Opens and edits memory files such as CLAUDE.md, letting you review project- or user-level memories quickly.

• /resume

  Commands

  Picks a previous session and resumes it. Useful to continue work you exited with Ctrl+C or hand off between machines.

• Agent

  Basics

  An autonomous unit of AI execution that iterates on tool calls and reasoning to accomplish a task. Claude Code itself is an agent, and it can spawn sub-agents to work in parallel.

• allowedTools

  Settings

  Part of the permissions block listing tool names that are allowed without prompting. You can match on tool + argument pattern, e.g. `Bash(git status)`.

• Anthropic

  Other & Troubleshooting

  The AI company behind the Claude model family and Claude Code. It combines safety-focused research with a commercial API business.

• apiKey

  Settings

  Credentials used to call the Anthropic API. You can also pass it via the `ANTHROPIC_API_KEY` environment variable, or skip it entirely when using Console login.

• autoEdit / Auto-accept

  Settings

  A mode that auto-accepts file edits and tool runs without prompting. Great for throughput, but safest when used on trusted repos inside a sandbox.

• Batch Processing

  Features

  Running Claude Code against many inputs at once, often by combining it with shell loops or `xargs` to review large numbers of files.

• Claude Code

  Basics

  Anthropic's terminal-based AI coding assistant. You can ask it in natural language to read, edit and run code, and it can operate directly on your project files.

  Related:

  • claude.com/claude-code

• Claude Haiku

  Other & Troubleshooting

  The lightweight Claude model, optimized for latency and price. Ideal for routine automation.

• Claude Opus

  Other & Troubleshooting

  The flagship Claude model, tuned for large codebases and tasks that demand heavy reasoning.

• Claude Sonnet

  Other & Troubleshooting

  The mid-tier Claude model, balancing cost and quality. Commonly used as Claude Code's default model.

• CLAUDE.md

  Basics

  A memory file placed at the project root or in your home directory. Conventions, architecture notes and dos/don'ts written here are loaded automatically at the start of every Claude Code session.

• Context

  Basics

  The full set of inputs the model can see at once, including chat history, system prompt and tool outputs. When it exceeds the limit, older information is compacted or dropped.

• Context Window

  Other & Troubleshooting

  The maximum number of tokens a model can process at once. The Claude 4 generation scales this to hundreds of thousands or even a million tokens, enabling whole-repo reasoning.

• deniedTools

  Settings

  The inverse of allowedTools: a list of tools/commands that must never run. Ideal for blanket-blocking dangerous ops like `Bash(rm -rf *)`.

• File Editing

  Features

  Claude Code's ability to read and write files directly in your repo. Destructive changes are gated by diff previews and permission prompts.

• Fine-tuning

  Other & Troubleshooting

  Retraining a pretrained model on extra data to adapt it to a specific task. Claude Code itself doesn't do this directly; prompt design and CLAUDE.md are the usual substitutes.

• Git Operations

  Features

  Working with git (branches, diffs, commits) through Claude Code. Unintended pushes are blocked by the permission system.

• Hooks

  Basics

  A mechanism for running arbitrary shell commands at Claude Code lifecycle events such as before/after tool use or on session start. Commonly used for auto-formatters, logging and permission checks.

• hooks (config)

  Settings

  The `hooks` key in settings.json defines shell commands to run for each lifecycle event, letting you build pre-commit-like custom flows into a project.

• Interactive Mode

  Features

  The default REPL-style mode when you launch `claude` without arguments. Best suited for iterative work and shared context.

• mcp (config)

  Settings

  The config block that registers MCP servers with their command, arguments and env vars. Claude Code starts them automatically at launch.

• MCP Server

  Features

  A process that exposes tools and resources over MCP. Registering it with Claude Code lets the assistant work with custom data sources and internal systems.

• model

  Settings

  Specifies which model to use (e.g. `claude-opus-4`, `claude-sonnet-4`, `claude-haiku-4`). Can also be overridden on the command line with `--model`.

• Model Context Protocol

  Basics

  An open protocol from Anthropic for connecting LLM clients to external tools and data sources. In Claude Code you can add MCP servers to extend the assistant with databases, issue trackers and more.

• One-shot Mode

  Features

  A non-interactive mode invoked with `claude -p "..."` that runs a single prompt. Ideal for scripts and CI pipelines.

• permissions

  Settings

  A top-level settings.json key that groups allow/deny rules for tool execution and persists the decisions you make during a session.

• Pipe Integration

  Features

  The ability to pipe another command's output into Claude Code's stdin, enabling one-liners such as `git diff | claude -p "review this"`.

• Prompt

  Basics

  The instruction sent to the model. In Claude Code the final prompt includes not only user input but also CLAUDE.md and the system prompt.

• Rate Limit

  Other & Troubleshooting

  A cap on requests or tokens per unit of time. Exceeding it returns a 429 error; you need to back off or upgrade your plan.

• REPL

  Other & Troubleshooting

  Short for Read-Eval-Print Loop: an interactive shell that evaluates inputs immediately. Claude Code's interactive mode is effectively a REPL.

• Session

  Basics

  A single continuous interaction from launching Claude Code to exiting. Each session has its own history, and you can resume a previous one with `/resume`.

• settings.json

  Settings

  Claude Code's configuration file. There are two levels—user-wide (~/.claude/settings.json) and per-project (.claude/settings.json)—with the project file taking precedence.

• Shell Execution

  Features

  Lets Claude Code run arbitrary shell commands on your behalf, making it easy to automate tests, builds and linters.

• Slash Commands

  Features

  Built-in or user-defined commands starting with `/`. You can also define custom ones per project under `.claude/commands/`.

• Streaming

  Features

  Receiving model output token by token so it renders in the terminal as it's generated, which dramatically reduces perceived latency.

• Terminal

  Other & Troubleshooting

  The text UI environment where Claude Code runs, such as macOS Terminal/iTerm2, Linux emulators, or Windows Terminal + WSL.

• Token

  Basics

  The smallest unit of text processed by an LLM. Roughly one word per token in English (and a higher ratio for CJK languages). Tokens are the basis for billing and context limits.

• Token (billing)

  Other & Troubleshooting

  Token as the unit used for billing and context accounting. Token counts depend on language, and CJK text typically costs more than equivalent English.

• Tool

  Basics

  A unit of capability Claude Code can invoke from the model's responses. File I/O, shell execution and web fetching ship as built-ins, and additional tools can be added through MCP.

• tools (config)

  Settings

  A setting that toggles which built-in tools are enabled. Disabling the ones you don't need improves both safety and latency.

No terms matched your search.