⌨️Claude Code
Anthropic's coding CLI: commands, settings and files · 528 terms
Claude Code Concepts
51The big ideas in Claude Code. How it runs, what it reads, and what it may do.
A coding agent from Anthropic. It reads, runs and edits the code in your project.
Where you use Claude Code: a terminal, an app or the web.
Sessions
6One conversation with Claude Code, how it is saved, and how it stays inside its space.
One saved chat with Claude Code, in one folder.
The file on disk where Claude Code keeps a session.
Your Claude Code chat that keeps going when the terminal is shut.
A save point Claude Code makes at each prompt, so you can go back.
Claude Code sums up an old chat to make room.
A note Claude Code slips into the chat to give Claude context.
Run Modes
6Ways to run Claude Code besides the terminal: in a script, in the cloud, or from a phone.
Running Claude Code on one prompt, with no chat, then exiting.
Starting Claude Code with none of your hooks, skills or memory loaded.
A Claude Code session that runs on a server and keeps going when you leave.
Drive the session on your computer from your phone.
Claude in its own copy of the code, on its own branch.
Several Claude Code sessions that work as one team.
Models
4How you pick the model, and how hard and how fast it works.
A short name like opus or sonnet that picks a model.
How hard the model thinks on each step, from low to max.
A word you add to one prompt so Claude thinks harder on that turn only.
The same model, but faster, at a higher price.
Instructions and Memory
5The files that tell Claude about your project, and the notes it keeps for itself.
A file of rules Claude Code reads each time it starts.
Your own CLAUDE.md for one project, kept out of git.
A line in CLAUDE.md that pulls in one more file.
Small rule files. Some load only for some files.
Notes Claude writes for itself, loaded again at the next session.
Extensions
13The pieces you add to Claude Code: skills, plugins, styles, a status line and helpers.
A folder with a SKILL.md file of steps. Claude Code loads it when a task needs it.
The header of a SKILL.md file. It tells Claude what the skill does.
Skills that ship with Claude Code, such as /code-review and /loop.
A package that adds skills, hooks, helpers and servers in one go.
The file that names and describes a Claude Code plugin. Only the name is required.
A list of plugins you can add, then install from.
Who gets a plugin: just you, your team, or you in one repo.
Rules for how Claude talks to you, for a whole session.
A bar under the prompt, filled by a script you choose.
Helpers that come with Claude Code, like Explore and Plan.
A helper agent you define in a Markdown file. Claude hands it tasks to do.
Loading an MCP tool's details only when Claude needs that tool.
A way for chat apps to send notes into a running session.
Permissions and Safety
11How Claude Code decides what it may do without asking, and what keeps it safe.
How often Claude Code stops to ask you before it acts.
Claude reads freely, but asks before it edits or runs.
Claude edits files on its own, but asks to run things.
Claude makes a plan and edits nothing until you say yes.
A second model checks each step. You are rarely asked.
Claude never asks. Anything not allowed up front is turned down at once.
Skips almost all checks. Use it in a throwaway box only.
A line that says yes, ask or no to a tool.
A wall that keeps Claude's shell commands in set folders and sites.
A yes or no that decides if a folder's settings are used.
Folders like .git and .claude that get an extra check.
Accounts and Usage
4How you sign in to Claude Code, who pays for it, how much you may use, and how a team tracks it.
How Claude Code signs you in. It picks one credential in a fixed order.
The caps on how much a Claude plan lets you use Claude Code before a reset.
Claude Code run through the Amazon, Google or Microsoft cloud. Your cloud account pays.
Sends Claude Code use, cost and tool data to your own dashboard.
Built-in Tools
38The things Claude Code lets Claude do, like read files, run commands and search the web.
One of the actions Claude Code gives Claude, like Read or Bash.
File Tools
7The tools that find, read and change the files in your project.
The tool Claude uses to open a file and see what is in it.
The tool that makes a new file or replaces a whole one.
The tool that swaps one exact piece of text in a file.
The tool that changes one cell in a notebook.
The tool that finds files whose names match a pattern.
The tool that searches inside files for a word or pattern.
The tool that gives Claude the code smarts of a language server.
Shell Tools
3The tools that run commands on your computer, now or in the background.
The tool that runs shell commands on your computer.
It runs PowerShell. Most of the time that is on Windows.
The tool that watches a command and tells Claude each new line.
Web Tools
2The tools that reach the web, to search it and to read a page.
The tool that reads one web page and pulls out what Claude asked for.
The tool that searches the web and returns titles and links.
Agent Tools
6The tools that start helpers, send them messages and stop them.
The tool that starts a subagent and gets back its answer.
It sends a note from one agent to another one.
The tool that lists the agents Claude can send messages to.
The tool that stops a running background task by its id.
How a helper hands back its report in auto mode.
It runs a script that sends out many helpers at once.
Session Tools
7The tools that steer the session itself: questions, plans, worktrees, skills and task lists.
The tool Claude uses to ask you a question with choices.
The two tools that move Claude into plan mode and back out.
Two tools that move a session into a worktree and back.
The tool that loads a skill's steps into the chat for Claude.
Four tools that let Claude keep a list of tasks and tick them off as it goes.
The old checklist tool. The Task tools took its place.
The tool that ends the session, used only in rare cases.
Schedule Tools
3The tools that run a prompt later, once or again and again.
Three tools that run a prompt at set times inside a session.
The tool Claude uses to choose when a self-paced loop runs next.
The tool that sets up scheduled tasks that run in the cloud.
MCP Tools
3The tools that reach MCP servers: find their tools, read their resources, wait for them.
The tool that loads a held-back tool when Claude needs it.
Two tools that list and read what MCP servers share.
It waits for a slow MCP server to be ready.
Output Tools
6The tools that hand work back to you, as a page, a file, an alert or a report.
It turns a file into a private web page.
It sends you an alert on your desktop or phone.
The tool that sends a file from the session to your device.
The tool that hands back code review findings as a list.
The tool that drafts a feedback report for you to send or not.
The tool that shares a team's onboarding guide as a link.
Integrations
14Ways to use Claude Code from the places your team already works, such as GitHub, GitLab and Slack.
Claude Code that runs as a job in a GitLab pipeline.
Mentioning @Claude in Slack to start a coding task in the cloud.
Claude Code on a GitHub a company runs on its own.
GitHub Actions
4Running Claude Code inside a GitHub workflow, and the parts that set it up.
Claude Code that runs in a GitHub workflow when you write @claude.
The app that lets Claude get into your GitHub repos.
Writing @claude in a GitHub comment to give Claude a task.
The settings that tell the GitHub Action what Claude should do.
Code Review
4The service that reviews pull requests on its own, and how to tune it.
A service where Claude checks each GitHub pull request on its own.
The three marks a Code Review gives each thing it finds.
A file that tells Code Review what to flag and what to skip.
A pull request comment that starts a Code Review now.
IDEs
3How to use Claude Code in the app where you write code. A dev container your editor opens counts too.
Claude Code in a side panel of VS Code. You see each change and accept it.
Links Claude Code to JetBrains IDEs. Changes open in the IDE's own diff viewer.
Claude Code running inside a dev container. Its commands stay in the box.
Commands
148What you type to run Claude Code: slash commands in a chat, and the claude command and its flags.
Slash Commands
71Commands you type after a slash to steer Claude Code itself, from a clean chat to its setup.
A word that starts with /, like /clear, that steers Claude Code.
Commands that start, name, save, measure or end the conversation you are in.
It starts a new chat with an empty context.
It brings back an earlier chat, with all of its history.
It copies the chat so far and moves you into the copy.
It gives the current session a name.
It shows what is filling the context window.
It sets when the chat is squeezed on its own.
It asks a quick side question that is not saved in the chat.
It sums up the session so far in one line.
It copies Claude's last reply to the clipboard.
It saves the whole chat as a plain text file.
It brings a cloud session to your terminal.
It closes Claude Code. The session is saved.
Commands that move work to the background, so it goes on while you do other things.
It sends this session to run in the back. Your terminal is free.
It copies this chat into a new session in the back.
It hands a task to a helper that reports back here.
It lists the background work in this session.
It ends the background session you are in.
Commands that pick the model, set a goal and steer the task Claude is doing.
It picks which model Claude Code uses.
It lets Claude ask a stronger model for advice at hard points.
It keeps Claude working until a goal you set is met.
It splits a big change into parts that run side by side.
It starts your app and uses it, to see a change work.
It checks a change works by running the app.
It lets the session work in one more folder.
It moves the session to a new folder. The chat stays.
It shows the cost and the limits you have left.
Commands that check your changes before you ship them, for bugs, cleanups and safety.
It shows each change in your files not yet committed.
It checks your changes for bugs before you ship them.
It looks at your changes for holes that let bad people in.
It tidies your changed code. It does not hunt for bugs.
It watches your pull request and pushes fixes for you.
Commands that set up a project, its rules and the tools Claude Code works with.
It writes a first CLAUDE.md by reading your project.
It opens the settings, or sets one from the prompt.
It shows and edits what Claude may do without asking.
It checks your setup and offers to fix what is wrong.
It brings your setup over from a tool you used before.
It hooks Claude Code up to the app you write code in.
It makes Shift+Enter give you a new line.
It walks you through using Claude Code on Bedrock or Google Cloud.
It links your GitHub account to cloud sessions.
It picks the default setup for your cloud sessions.
It drafts the list of places auto mode may trust.
It finds safe commands you keep saying yes to, and allows them.
You say a settings change, and Claude edits the file.
Commands for your account, your plan and the version you are running.
It signs you in to your Anthropic account. /logout signs you out.
It shows what you run, and who you are signed in as.
It opens the page to move to a bigger plan.
It shows and changes your privacy choices.
It restarts Claude Code to apply an update, and keeps the chat.
Commands that change how Claude Code looks, and how you type into it.
It changes the colors of Claude Code.
It switches between the classic and fullscreen views.
It hides the steps and shows your prompt and the answer.
It colors this session's prompt bar.
It opens the file where you set which keys do what.
It lets you speak your prompt instead of typing it.
Commands for when you are stuck, or something is not working the way it should.
It lists the commands you can use, with help.
It sends a bug report or feedback about Claude Code.
It turns on the debug log and reads it to find a problem.
It makes a report on how you use Claude Code.
It shows what changed in each version of Claude Code.
Commands for the parts you add to Claude Code. These are MCP servers, hooks, skills, plugins, the browser and workflows.
It shows your MCP servers and lets you fix their links.
It lists your hooks and where each one comes from.
It lists your skills and lets you hide the ones you do not want.
It shows what each skill costs, and how often it is used.
It finds skills you added during the session, with no restart.
It brings in plugin changes, with no restart.
It sets up and checks Claude's link to your browser.
It adds the Claude app to your Slack workspace.
It shows running workflows and lets you pause or save them.
It lists your artifact pages so you can open or reuse one.
CLI Commands
29Commands you type in the shell, starting with claude and a word, such as claude update.
A command you type in the shell after claude, outside a chat.
Commands that install, update, check and clean up Claude Code on your computer.
It moves Claude Code to the newest version.
It installs Claude Code, at the version you pick.
It checks your install from the shell and changes nothing.
It deletes what Claude Code keeps for one project.
It shows auto mode's rules, and resets your changes to them.
Commands that sign you in, show who you are signed in as, and make tokens for scripts.
It signs you in from the shell. claude auth logout signs you out.
It says if you are signed in, as JSON or text.
It makes a sign-in token for CI and scripts. It lasts one year.
Commands for jobs that run in the background. You use them to look at, open, restart and remove them.
It opens a job that runs in the background, here.
It prints what a background job did just now.
It restarts a background session and keeps its chat.
It takes a background session off the list.
It checks and stops the process that runs background jobs.
Commands that add, list, remove and sign in to MCP servers from the shell.
It adds an MCP server to Claude Code from the shell.
It adds an MCP server from a block of JSON.
It lists your MCP servers and whether each one works.
It takes an MCP server out of Claude Code.
It signs you in to one MCP server from the shell.
It runs Claude Code as an MCP server for other apps.
Commands that install, list and remove plugins, and check a plugin you make.
It installs a plugin from the shell.
It removes a plugin, and prune clears what nothing needs.
It lists installed plugins with their version and scope.
It adds a list of plugins you can install from.
It checks a plugin you are making for mistakes.
It tests how much a plugin helps, with scored cases.
Commands that run Claude Code as a server, for a phone, a company or the cloud.
It lets you drive Claude Code on your computer from your phone.
It lets a firm add its own sign-in and rules.
It lets cloud sessions run on your own machines.
CLI Flags
48Flags you add after claude to change how one session starts, without changing your settings.
Flags that find a session again, by its name, its ID or its pull request.
It opens an old session so you can keep going.
It gives a new session a name you choose.
It starts a session with an ID you choose.
It lists the sessions that worked on one pull request.
Flags that pick where a session runs, such as an app, your editor or your own machines.
It starts a session in the background and hands you back the shell.
It opens this folder in the Claude Desktop app.
It starts a cloud session on your own machines.
It links the session to your code editor at the start.
It opens a worktree session in its own tmux panes.
It sets how agent team members show on your screen.
Flags for claude -p that set the shape of the answer and the limits of the run.
It picks the shape of what claude -p prints.
It makes claude -p answer in a JSON shape you set.
It caps how many turns claude -p may take.
It stops claude -p when the spend hits a dollar cap.
It stops claude -p from saving the session to disk.
It says who answers the yes or no questions in print mode.
Flags that add more kinds of events to a stream-json run, for apps that follow it live.
It sets how claude -p reads what you send it.
It streams the answer in small pieces as Claude writes it.
It adds hook events to the output stream.
It adds what subagents write to the output stream.
It adds a guess at the next prompt to the stream.
It shows the full output of each turn.
Flags that set which tools Claude has, which run without asking, and who answers when one asks.
It sets which built-in tools Claude has in a session.
It lists tools that run without asking you first.
It lists tools, or uses of them, that are not allowed.
It names an MCP tool that says yes or no for you.
It lets you switch to bypass mode later on.
Flags that set what a session loads when it starts: its settings, its add-ons and its model.
It loads extra settings for one session from JSON.
It picks which settings files Claude Code reads.
It starts Claude Code with your add-ons turned off.
It starts a locked session for a shared box.
It names backup models for when the main one is busy.
It turns on API features that are still in testing.
It prints plain text that screen readers can read well.
It prints which version of Claude Code you have.
Flags that replace or add to the system prompt Claude Code sends with every request.
It swaps the whole default prompt for yours.
It adds your text to the end of the default system prompt.
It says if a chat keeps the system prompt it started with.
It makes the prompt the same for each user.
It adds your text to the prompt of each subagent.
Flags that load agents, MCP servers, plugins and setup hooks for one session.
It runs the main session as one of your custom agents.
It sets up helper agents for one session.
It loads MCP servers for one session from JSON.
It loads only the MCP servers you pass with --mcp-config.
It loads a plugin from a folder for one session.
It turns off all skills and slash commands for a session.
It turns on channels that are not approved yet, for testing.
It runs your Setup hooks before a session starts.
Configuration
252How you set up Claude Code. It covers the values and keys it reads, the rules it must obey, the scripts it runs and the files that hold them.
Environment Variables
130Variables that change how Claude Code signs in, connects and runs. You set them in the shell or in a settings file.
Variables that hand Claude Code a key or token to sign in with.
It gives Claude Code an API key to sign in and pay with.
It sends a token that a gateway can check.
It lets claude auth login sign in with no browser.
It names the saved profile to sign in with.
These set the path a request takes. They pick the gateway, the proxy and the certificates.
It sends the API requests to a new address.
It adds your own headers to each request.
It turns on API betas for each request.
It gives Claude Code a certificate to prove who it is.
It picks which sets of certificates Claude Code trusts.
Set to 0, it stops Claude Code from squeezing its requests.
Set to 1, it lets the proxy look up host names.
Variables that pick which model Claude Code runs on, and what each short model name means.
It sets the model each new session starts on.
It sets the model to use when nothing else picks one.
It sets which full model the alias opus stands for.
It sets the name the model picker shows for a pinned model.
It tells Claude Code which features a pinned model has.
It adds one model of your own to the /model picker.
It sets the model for helpers that have none.
Set to 1, it puts one model on each helper.
Set to 1, it keeps old Opus 4 names from moving to the new Opus.
Set to 1, it holds every model to a 200K window.
Set to 1, it turns fast mode off.
It picks the model that checks actions in auto mode.
Set to 1, it fails a request in place of a quiet model switch.
These cap how much Claude Code reads, writes and starts. That means tokens, characters, memory and helpers.
It sets how many tokens Claude may write in one reply.
It sets how many tokens Claude may spend thinking.
It sets how much of a file one read can return.
It sets how big an MCP tool's answer may be.
It sets how much of a command's output Claude gets back.
It sets how much skill text Claude is shown.
It sets how full a chat gets before it is squeezed.
It sets the percent full at which a chat is squeezed.
It tells Claude Code the model's real context size.
It sets how many subagents can run at the same time.
It sets how deep helpers may start helpers.
It caps how many web searches one session can run.
It caps how much memory Claude's commands can use.
These set how long Claude Code waits for a command, a request or a server. Then it gives up.
It sets how long a command runs before it is stopped.
It sets the longest time Claude may ask a command to run.
It sets how long an API request waits for its reply to start.
It sets how many times a failed request is tried again.
It sets how long Claude Code waits for an MCP server to start.
It sets how long one MCP tool call may run.
It sets how long a file search may run, in seconds.
These run Claude Code on a cloud such as Bedrock, Google Cloud or Foundry. They pick the cloud, the region and the key.
Set to 1, it runs Claude Code on Amazon Bedrock.
It skips the cloud sign-in when a gateway signs in for you.
It gives Claude Code a Bedrock API key to sign in with.
It picks which group of Bedrock regions to try first.
It picks how fast and how cheap Bedrock serves you.
It sets a second AWS region for the small model.
Set to 1, it makes Claude Code look up AWS keys on every request.
It sets the Google Cloud region Claude Code uses.
It sets the Google Cloud region for one model.
It names the Google Cloud project Claude Code bills to.
It names the Foundry resource Claude Code uses.
It gives Claude Code a key to sign in to Microsoft Foundry.
Set to 1, it fills the model picker from your gateway's list.
An app that runs Claude Code sets it to keep control of routing.
These set how hard Claude thinks: the effort level, the thinking field and one key word.
It sets the effort level, from low to max.
Set to 1, it sends effort even for a model name it does not know.
Set to 1, it leaves the thinking field out of each request.
Set to 1, it gives two old models a fixed thinking budget.
Set to 1, it stops thinking between tool calls.
Set to 1, it turns off the ultrathink keyword.
These set prompt caching: whether it is on, and how long a saved prompt lasts.
Set to 1, it turns prompt caching off for every model.
Set to 1, it keeps the prompt cache for an hour.
It sets the cache time for the main chat: 5m or 1h.
Set to 1, it makes the cache last five minutes, every time.
These change what goes in each request, for a gateway that needs it.
It adds your own fields to each model request.
Set to 1, it strips pre-release betas from each request.
Set to 1, it stops the one-piece retry when a stream breaks.
It sets how many tries the model gets to match a schema.
These decide what goes into the chat for Claude to read: notes, @ files and the squeeze.
Set to 1, it stops the automatic squeeze. /compact still works.
Set to 1, it turns off all compaction, /compact too.
Set to 1, it loads no CLAUDE.md or memory files.
Set to 1, it turns auto memory off. Set to 0, it forces it on.
Set to 1, it loads CLAUDE.md from folders added with --add-dir.
Set to 1, it drops the built-in git steps from the chat.
Set to 1, an @ mention is sent as text, not as the file.
These turn a Claude Code feature on or off, such as updates, a tool, or a background job.
Set to 1, it stops updates in the background. Updating by hand still works.
Set to 1, it blocks every update, even by hand.
Set to 1, it saves no copies. Then /rewind cannot undo edits.
Set to 1, it keeps each command and helper in front.
Set to 1, it removes the skills that ship with Claude Code.
Set to 1, it turns off timed tasks in a session.
Set to 1, it turns off WebFetch. Web search still works.
Set to 1, it turns workflows off.
Set to 1, it turns off the tool that posts web pages.
Set to 1, it turns off the Explore and Plan subagents.
Set to 1, it turns on the agent teams test feature.
Set to false, it turns off the gray prompt guesses.
It forces the come-back recap on with 1 or off with 0.
Set to 1, it hides the cost warning messages.
These decide what Claude Code reports, home or to your own tools, and what goes in its logs.
Set to 1, it sends metrics and logs to your own collector.
Any value stops the usage data sent home.
Any value stops crash reports from going home.
Any value cuts each call Claude Code can do without.
Set to 1, it puts your prompt text into telemetry logs.
Set to 1, it adds tool inputs and names to telemetry.
It logs the full JSON of each model request and answer.
Set to 1, it stops the How is Claude doing? question.
Set to 1, it turns on debug logs, like --debug.
These change how Claude Code draws in your terminal: the screen, the mouse, colors and links.
Set to 1, it turns on fullscreen mode, with less flicker.
Set to 1, it keeps the classic mode, with the terminal's own scroll.
Set to 1, fullscreen mode leaves the mouse to your terminal.
It sets how far the mouse wheel scrolls in fullscreen mode.
Set to 1, fullscreen draws each message, not just the ones in view.
Set to 1, it keeps the real cursor, so zoom tools can find it.
Set to 1, it leaves your terminal's title alone.
Set to false, it turns off the colors in diffs.
Any value lets tmux show full color.
Set to 1, it turns on links you can click in the terminal.
Set to 1, it hides the folder path in the start logo.
These change the shell that runs Claude Code's commands, and what that shell sees.
It picks bash or zsh to run Claude Code's commands.
It wraps each shell command in a tool of yours, often a log.
On Windows, it points Claude Code at your Git Bash.
A script that runs before each Bash command, so its exports stay.
It goes back to the start folder after each Bash command.
Set to 1, it hides your secrets from the commands Claude Code runs.
Set to 0, it searches with your own rg, not the built-in one.
It moves the folder where Claude Code keeps scratch files.
These name the session and its folders, or tell a script it is running inside Claude Code.
It moves Claude Code's own folder, ~/.claude by default.
Set to 1 in each program Claude Code starts, so a script knows.
It holds the current session's id, for scripts and hooks.
It holds Claude Code's own process id, for the scripts it starts.
Set to true when Claude Code runs as a cloud session.
Give sessions the same id, and they share one task list.
It moves the whole plugins folder, not just the cache.
Set to false, it skips the MCP servers you added on claude.ai.
Set to 0, it waits for MCP servers before the first prompt.
Permission Rules
16The rules that say which tools Claude Code may use without asking, and how each kind matches.
One kind of rule per tool. Shell commands, files, web pages, MCP servers and helpers each have one.
A rule like Bash(npm run *) that covers shell commands.
A rule that stops Claude from reading a file, like your keys.
A rule that says which files Claude may change.
A rule that says which web sites Claude may fetch.
A rule that covers the tools of one MCP server.
A rule that turns one helper agent on or off.
A deny or ask rule that checks one value a tool gets.
How a rule decides if it covers a call: the star, joined commands, paths and the prompt.
A star in a rule that stands for any text, and where it goes matters.
Commands joined with && or a pipe, each part checked on its own.
A small program that runs a command, and that rules look past.
Shell commands that only look, so they never ask.
The path in a Read or Edit rule. How it starts sets where it points.
The question Claude Code asks when no rule allows a call.
Which folders Claude Code may read and edit, and how to add or fence off more.
The folder you start Claude Code in, where it reads freely.
More folders Claude may read and edit, beside the first one.
A setting that keeps the file tools in your work folders.
Hooks
35Scripts Claude Code runs on its own at fixed points, such as before a tool runs or when Claude stops.
A script Claude Code runs on its own at a fixed point, such as before a tool.
Events around each tool call. They fire before it, after it, when it fails, and when it needs your yes.
The event before a tool runs. A hook can say yes, no or ask.
The event right after a tool works. It is often used to tidy code.
The event when a tool fails, used to log, alert or give Claude a hint.
The event just before Claude Code asks you. A hook can answer.
The event when auto mode refuses a tool call.
Events that fire once a turn: when you send a prompt, when Claude stops, and when a helper starts or ends.
The event when you send a prompt, before Claude reads it.
The event when Claude finishes, where a hook can make it keep going.
The event when a turn ends on an error. It is for logs and alerts.
The event when Claude Code needs you, used for desktop or phone pings.
The event when Claude starts a helper agent.
The event when a helper agent ends. A hook can make it go on.
How a hook answers back: its exit code, the JSON it prints, and text it adds for Claude.
The number a hook ends with: 0 goes ahead, 2 blocks, others warn.
A JSON reply from a hook. It can say more than an exit code.
A hook field that adds a note for Claude to read.
Events tied to the session itself. They fire when it starts or ends, when the chat is cut down, and when files or folders change.
The event when a session starts. It can load notes for Claude.
The event when a session ends. It is used to clean up or save logs.
An event that fires only on a set-up run.
The event just before the chat is cut down. A hook can save it first.
The event after the chat is cut down. A hook can log the summary.
The event when a rule file is loaded for Claude.
The event when a settings or skill file changes.
The event when Claude moves to a new folder with cd.
The event when a watched file changes, no matter what changed it.
The kinds of handler a hook can run. They are a command, a web request, an MCP tool, a prompt or an agent.
A hook that runs a shell command and reads JSON from its input.
A hook that sends the event to a web server.
A hook that calls a tool on an MCP server.
A hook that asks a Claude model a yes or no question.
A hook that sends a helper to look at files. Then it answers.
The fields and paths that decide when a hook runs, how long it may take, and where its script lives.
The field that picks which tools or events a hook runs on.
A field that runs a hook only when the call matches one rule.
A command hook that runs in the background while Claude goes on.
How many seconds a hook may run before Claude Code stops it.
The path to the top folder of a project, for hook scripts.
The path to the folder a plugin is in, for its hooks.
Config Files
12The files Claude Code reads its setup from, in a project and in your home folder.
A JSON file that sets how Claude Code acts. It holds its rules.
The files of settings at each level, from yours to your company's. Also the order that says which one wins.
Your own file of settings. It works in each repo you open.
A settings file in the repo. The whole team shares it in git.
Your own settings for one project, kept out of git.
A file of settings that a company puts on each machine. It beats yours.
The order that says which file wins when two set the same key.
Files that live in a project and travel with it in git.
A folder in a repo that holds its Claude Code files.
A file in a project that lists MCP servers the team shares.
The list of files a new worktree needs, like .env.
Files in your home folder. They hold your own setup for each repo.
The .claude folder in your home folder, for your own setup.
A file Claude Code writes for itself. It holds your account and MCP servers.
The file where you change Claude Code's keyboard shortcuts.
Settings
59The keys you write in a settings file, grouped by what they change.
Keys that pick the model, limit which ones may be used, and set how hard it thinks.
The key that sets which model a new session starts on.
A list that limits which models people may pick.
The key that sets how hard Claude thinks by default.
A list of backup models for when the main one is busy.
The key that can switch off thinking in all your chats.
The key that sets which language Claude replies in.
Keys that set the permission mode a session starts in, and which modes are allowed at all.
The key that sets which permission mode a new session starts in.
The key that stops anyone from turning on the skip-all mode.
The key that says no to a call instead of waiting for you.
The key that takes auto mode out of the list of modes.
Keys for the session itself: its variables, its key, its logs, and what it writes to git.
The key that sets shell values for all your chats.
A command that hands Claude Code a fresh API key.
The key that sets how many days old chat logs are kept.
The key that can leave out Claude Code's own git help.
The key that changes or hides Claude's credit line in commits.
The key that sets the folder where plan mode saves its plans.
Keys for what you see and touch: the screen, the box you type in, and the lines that show progress.
Keys that change how the screen looks: the colors, the view, how much of each turn you see, and the motion.
The key that picks the colors Claude Code shows.
The key that picks the classic or the full-screen view.
The key that sets how much of each turn you see.
The key that turns off the moving parts of the screen.
The key that shows code as plain text, with no colors.
Keys for the box you type in: its keys, its file picker, its shell and your voice.
The key that turns on vim-style keys in the input box.
The key that lets two letters in a row, such as jj, act as Esc in vim mode.
A command of your own that suggests files when you type @.
The key that hides git-ignored files from the @ picker.
The key that picks Bash or PowerShell for your ! commands.
The key that turns on voice typing and sets the talk key.
Keys that show what Claude is up to: the status line, the spinner, the clock and the tab name.
The key that sets the command behind the status line.
The key that changes the words the spinner shows.
The key that hides the tips shown while Claude works.
The key that shows how long each turn took.
The key that decides if the tab shows the name you gave.
Keys IT sets in managed settings, so people cannot change them on their own.
Keys that stop people adding hooks, rules, skills and plugins from their own files.
The key that lets only the company's hooks run.
The key that lets only IT write the rules for what Claude may do.
The key that lets add-ons come only from plugins.
The key that lists which plugin stores people may add and use.
Keys that list which MCP servers may load and which are blocked.
A list of the only MCP servers people may add.
A list of MCP servers that may never load.
The key that makes IT's MCP allow list the only one.
Keys for who may log in, which version may run, which models are blocked, and what shows at the start.
The key that picks which kind of account people log in with.
The key that keeps logins inside one work account.
The key that stops old versions of Claude Code from starting.
A list of models no one may use, set by the company.
The key that shows company messages when Claude Code starts.
A program IT runs to work out the rules each time Claude Code starts.
Keys under sandbox that fence in the Bash commands Claude runs: which ones, which files and which hosts.
Keys that turn the sandbox on and decide which commands run in it, and if you are asked.
The key that runs Claude's shell commands inside a sandbox.
The key that skips the prompt for commands in the sandbox.
The key that says if Claude may run a blocked command with no fence.
A list of commands that run outside the sandbox.
The key that stops Claude Code if the sandbox cannot start.
Keys for the files a fenced command may read and write, and the secrets it may not see.
A list of more folders that fenced commands may write to.
A list of paths sandboxed commands may never write to.
A list of paths sandboxed commands may not read.
A list of paths that may be read inside a blocked area.
The key that hides or masks secrets from sandboxed commands.
Keys for the hosts and ports a sandboxed command may reach.
A list of web hosts that commands in the sandbox may reach with no prompt.
A list of hosts sandboxed commands may never reach.
The key that lets fenced commands on a Mac start a server.
The key that sends sandbox traffic through your own proxy.
Keyboard Shortcuts
25The keys you press in Claude Code to stop, send, switch, view and type faster.
Control Shortcuts
7Keys that stop a turn, quit, send a queued note now, or move work to the background.
The key that stops a task, clears the prompt, or quits Claude Code.
The key that stops Claude in the middle of a turn and keeps the work done so far.
Press Esc twice to clear the prompt, or to go back to an earlier point.
The key that quits Claude Code when you press it twice, fast.
The key that lets a long task run in the back while you keep chatting.
Two keys, pressed in turn, that stop all the subagents that run in the back.
The key that sends what you queued now, and does not wait for the turn to end.
Mode Shortcuts
4Keys that switch the mode, the model, thinking, or fast mode in one press.
The key that moves Claude Code to the next permission mode.
The key that opens the list of models and keeps what you typed in the prompt.
The key that turns extended thinking on or off for the model.
The key that turns fast mode on or off. Answers come faster, and cost more.
Prompt Shortcuts
6Keys for the box you type in: an editor, a stash, an image, your past prompts and new lines.
The key that lets you write your prompt in your own editor.
The key that puts what you typed aside, and brings it back when you want it.
The key that pastes an image from the clipboard into your prompt.
The key that searches your past prompts by a word you remember.
The key that adds a new line to the prompt instead of sending it.
The key that brings back your past prompts and drafts, newest first.
Quick Commands
4One key at the start of the prompt that runs, finds, helps, or adds an emoji.
The ! prefix that runs your own shell command and puts its output in the chat.
Typing @ in the prompt to pick a file by name and point Claude at it.
Typing ? on an empty prompt to list the keyboard shortcuts.
Put a word between two colons to add an emoji. A heart is :heart:.
View Shortcuts
4Keys that change what the screen shows: the full transcript, the to-do list, a clean redraw and your voice.
The key that opens the whole log, with each tool call shown in full.
The key that shows or hides Claude's to-do checklist for the task.
The key that redraws a broken screen and keeps your chat and prompt.
Holding the Space key to speak your prompt instead of typing it.