Add-ons

Claude Code Cheat Sheet

Checked 2026-10-31

Loading poll...

This is the short list of Claude Code commands, flags, keys and file paths we reach for, each one checked against Anthropic's documentation on 31 October 2026. Start a session with claude, run one question with claude -p "...", continue with claude -c, stop with Esc, cycle modes with Shift+Tab, and manage extras with /mcp, /skills and /plugin.

Quick facts


Start a session	`claude`
One-shot, then exit	`claude -p "query"`
Continue last conversation	`claude -c`
Resume by name or ID	`claude -r "<session>"`
Update	`claude update`
Health check	`claude doctor` (terminal) or `/doctor` (in session)
Stop the model	`Esc`
Cycle permission mode	`Shift+Tab`
Add an MCP server	`claude mcp add`

Start and run

Command	What it does
`claude`	Interactive session
`claude "explain this project"`	Interactive session with a first prompt
`claude -p "explain this function"`	Answer through the SDK, then exit
`cat logs.txt \| claude -p "explain"`	Pipe content in
`claude -c`	Continue the most recent conversation in this directory
`claude -c -p "Check for type errors"`	Continue, non-interactively
`claude -r "auth-refactor" "Finish this PR"`	Resume a named session with a new prompt
`claude update`	Update to the latest version
`claude install [version]`	Install or reinstall the native binary; accepts a version, `stable` or `latest`
`claude doctor`	Read-only install and settings diagnostics without starting a session
`claude auth login`, `claude auth logout`, `claude auth status`	Sign in, sign out, show status as JSON
`claude setup-token`	Make a long-lived OAuth token for CI and scripts. Needs a Claude subscription

Flags you will use

claude --help does not list every flag, so an absence there does not mean a flag is gone.

Flag	What it does
`--model <alias>`	Pick the model for this session, such as `sonnet`
`--effort <level>`	Set effort: `low`, `medium`, `high`, `xhigh`, `max`
`--permission-mode <mode>`	Start in `default`, `acceptEdits`, `plan`, `auto`, `dontAsk` or `bypassPermissions`
`--add-dir <path>`	Add another working directory
`--allowedTools`, `--disallowedTools`	Tools that run without a prompt, and tools to remove
`--tools "Bash,Edit,Read"`	Restrict the built-in tool set. `""` disables all
`--continue`, `-c` / `--resume`, `-r`	Continue or resume
`--name`, `-n`	Name the session for the resume picker
`--fork-session`	When resuming, start a new session ID
`--worktree`, `-w`	Run in an isolated git worktree under `.claude/worktrees/`
`--agents '<json>'` / `--agent <name>`	Define subagents for a session / run as a given agent
`--mcp-config <file>`	Load MCP servers from JSON files or strings
`--append-system-prompt "..."`	Add text to the default system prompt
`--system-prompt "..."`	Replace the whole system prompt
`--settings <file or json>`	Override settings for this session (file up to 2 MiB)
`--bare`	Minimal mode: skip auto-discovery of hooks, skills, commands, subagents, plugins, MCP servers, auto memory and CLAUDE.md
`--bg`, `--background`	Start as a background agent and return
`--debug`, `--debug-file <path>`	Debug logging, optionally filtered, such as `--debug='mcp,startup'`
`--verbose`	Full turn-by-turn output
`--version`, `-v`	Print the version

Print mode only (`-p`)

Flag	What it does
`--output-format text\|json\|stream-json`	Choose the output shape
`--input-format text\|stream-json`	Choose the input shape
`--json-schema '<schema>'`	Get validated JSON output matching a schema
`--max-turns <n>`	Limit agentic turns; exits with an error at the limit
`--max-budget-usd <n>`	Stop once estimated API spend reaches this amount
`--no-session-persistence`	Do not save the session to disk

One flag to treat with care: --dangerously-skip-permissions is equivalent to --permission-mode bypassPermissions. The name is the warning. Use it only in a sandbox or container you can throw away.

Keyboard shortcuts

On macOS, the Alt shortcuts (Alt+B, Alt+F, Alt+D, Alt+Y, Alt+P) need Option set as Meta in your terminal.

General

Keys	Action
`Esc`	Stop the current response so you can redirect; Claude keeps work done so far
`Esc` `Esc`	Clear the input draft (saved to history), or open rewind when empty
`Ctrl+C`	Interrupt; with nothing running, first press clears input, second exits
`Ctrl+D`	Exit (press twice)
`Shift+Tab`	Cycle permission modes: default, acceptEdits, plan, then bypassPermissions and auto when available
`Option+P` / `Alt+P`	Switch model without clearing the prompt
`Option+T` / `Alt+T`	Toggle extended thinking
`Option+O` / `Alt+O`	Toggle fast mode
`Ctrl+O`	Transcript viewer: detailed tool use, with timestamps and the model per message
`Ctrl+R`	Reverse-search your command history
`Ctrl+B`	Background running commands and agents (tmux users press twice)
`Ctrl+T`	Show or hide Claude's task checklist
`Ctrl+G` or `Ctrl+X Ctrl+E`	Edit your prompt in your editor
`Ctrl+V` (or `Cmd+V` in iTerm2, `Alt+V` on Windows and WSL)	Paste an image
`Ctrl+S`	Stash the prompt; press again on an empty prompt to restore it
`Ctrl+L`	Redraw the screen
`Ctrl+Z`	Suspend (Unix only); `fg` to resume
`Ctrl+X Ctrl+K`	Stop all running background subagents

Editing the prompt

Keys	Action
`Ctrl+A` / `Ctrl+E`	Start or end of line
`Alt+B` / `Alt+F`	Back or forward one word
`Ctrl+K`	Delete to end of line
`Ctrl+U`	Delete to start of line
`Ctrl+W`	Delete back to the previous whitespace; one press removes a whole path
`Alt+D`	Delete to end of word
`Ctrl+Y`, then `Alt+Y`	Paste deleted text, then cycle through earlier deletions
`Ctrl+_`	Undo the last input edit
`Up` / `Down`	Move through history

Multiline input

Option+Enter (after enabling Option as Meta on macOS) and Ctrl+J start a new line. Ctrl+J works in any terminal without configuration. Run /terminal-setup to install a Shift+Enter binding in VS Code, Cursor and similar terminals.

Slash commands

The full list is in our slash commands guide. These are the ones for daily use.

Command	What it does
`/help`	List commands
`/clear`	New conversation, empty context
`/compact [focus]`	Summarise to free context
`/context`	Show what fills the context window
`/model`	Switch and save the model
`/effort`	Set effort level
`/plan [task]`	Enter plan mode
`/permissions`	Manage tool rules
`/resume`	Pick a past conversation
`/rewind`	Roll back conversation, code, or both
`/diff`	Review working-tree changes
`/code-review` (alias `/review`)	Review a PR, branch or path
`/security-review`	Check the branch diff for vulnerabilities
`/init`	Create a CLAUDE.md
`/memory`	Edit CLAUDE.md files and auto memory
`/mcp`	MCP connections and OAuth
`/skills`, `/skill-doctor`	List skills; see what each costs in context
`/plugin`	Manage plugins
`/hooks`	See hook configuration
`/usage`	Cost, limits, stats (`/cost` and `/stats` are aliases)
`/doctor`	Setup checkup
`/loop [interval] [prompt]`	Repeat a prompt while the session is open
`/background`, `/fork`, `/tasks`	Run work in the background and manage it

A command is recognised only at the start of your message. Up to six skills can be chained in one message: /skill-a /skill-b do XYZ.

The .claude folder

Project files live in your repo; global ones live in ~/.claude/.

File	Scope	Commit it	What it does
`CLAUDE.md`	Project and global	Yes	Instructions loaded every session
`rules/*.md`	Project and global	Yes	Topic-scoped instructions, optionally path-gated
`settings.json`	Project and global	Yes	Permissions, hooks, env vars, model defaults
`settings.local.json`	Project only	No	Your personal overrides
`.mcp.json`	Project only (repo root)	Yes	Team-shared MCP servers
`skills/<name>/SKILL.md`	Project and global	Yes	Reusable prompts, `/name` or auto
`commands/*.md`	Project and global	Yes	Single-file prompts; same mechanism as skills
`agents/*.md`	Project and global	Yes	Subagent definitions
`output-styles/*.md`	Project and global	Yes	Custom instruction sets
`keybindings.json`	Global only	n/a	Custom keyboard shortcuts
`~/.claude.json`	Global only	No	App state, account details, personal MCP servers

Precedence, per the docs: managed settings beat every settings file, and CLI flags such as --permission-mode or --settings beat settings.json for that session.

MCP in one screen

# remote server
claude mcp add --transport http <name> <url>
# with a header
claude mcp add --transport http secure-api https://api.example.com/mcp \
  --header "Authorization: Bearer your-token"
# local server (the -- separates Claude's flags from the server's)
claude mcp add --transport stdio myserver -- npx server
# with an environment variable
claude mcp add --env KEY=value --transport stdio myserver -- python server.py
# from JSON
claude mcp add-json example '{"command":"npx","args":["-y","@example/mcp-server"]}'

Scope flag	Stored in	Shared
`--scope local` (default)	`~/.claude.json`, this project only	No
`--scope project`	`.mcp.json` in the repo	Yes, with an approval prompt
`--scope user`	`~/.claude.json`, all projects	No

Check servers with /mcp in a session. Sign in to an OAuth server from the shell with claude mcp login <name>. The SSE transport is deprecated; use HTTP where you can. For worked setups, see Playwright MCP, GitHub MCP and Context7 MCP, or the Claude Code MCP servers guide.

Skills, subagents and plugins in one screen

~/.claude/skills/<name>/SKILL.md     personal skill, all projects
.claude/skills/<name>/SKILL.md       project skill, commit it
~/.claude/agents/<name>.md           personal subagent
.claude/agents/<name>.md             project subagent
/plugin marketplace add <owner/repo> add a plugin marketplace
/plugin install <plugin>@<market>    install a plugin

Skill frontmatter you will reach for: description, disable-model-invocation: true, allowed-tools, argument-hint. Subagent frontmatter: name and description required, then tools, model, permissionMode, skills. Guides: skills, subagents, plugins.

Recipes

Review a branch from CI-style scripts

git diff main...HEAD | claude -p "Review this diff for bugs and security issues. List findings by severity." --output-format json

Cap a scripted run

claude -p "Fix the failing tests" --max-turns 10 --max-budget-usd 2

Try a risky change in isolation

claude --worktree try-new-parser

Read-only exploration

claude --permission-mode plan

Resume the work you left yesterday

claude -c
claude -r "auth-refactor"

Run with only the tools you choose

claude --tools "Read,Grep,Glob"

A starter settings.json

The folder table says settings.json holds permissions, hooks and environment variables. A small project file removes most repeat prompts without opening the door wide. This is our suggested starting point, so edit the commands to match your project, and check the permissions page for the rule syntax your version supports.

{
  "permissions": {
    "allow": [
      "Bash(npm test:*)",
      "Bash(npm run lint:*)",
      "Bash(git status)",
      "Bash(git diff:*)"
    ],
    "deny": [
      "Read(./.env)",
      "Read(./.env.*)",
      "Bash(rm -rf:*)"
    ]
  }
}

Commit it so teammates get the same rules, and keep personal changes in settings.local.json, which stays out of git. The allow list covers commands you run constantly and that cannot do harm, and the deny list protects secrets and the most destructive command. A deny rule is the right tool for "never", because instructions in CLAUDE.md are advice the model can miss, while a permission rule is enforced by the program. Add to the allow list when a prompt annoys you three times, not before.

Which permission mode when

Shift+Tab cycles the permission mode, and --permission-mode sets it at launch. The right mode depends on how much you trust the task and the environment.

Situation	Mode to pick	Why
Exploring code you do not know	Plan	Read-only, so nothing changes while you learn
A small, reviewed change on a branch	Default, with a short allow list	You see each edit and each risky command
Repetitive edits across many files	Accept edits	Saves clicking, and the diff is still reviewed in git
A throwaway container with no secrets	The most permissive mode your policy allows	Nothing to protect, and no one to answer prompts
A shared or production machine	Default or Plan only	The cost of a mistake is high

Whatever mode you use, the real safety net is version control. Commit before a long run, so any result can be reviewed as a diff and any mistake reverted in one command.

Cost and context habits

Sessions get more expensive as they grow, because each turn carries the history. A few commands from the tables above keep that in check.

Habit	Command	Effect
Check what fills context	`/context`	Shows memory files, tools and history as shares of the window
Clear between unrelated tasks	`/clear`	Starts clean, so old tasks stop costing tokens
Shrink a long session	`/compact` with a focus note	Keeps the thread and drops the detail
See spend or plan usage	`/usage`	Shows where the budget went
Use a cheaper model for routine work	`/model`	Saves the larger model for hard problems
Cap a scripted job	`--max-turns` and `--max-budget-usd`	Stops a stuck run at a known cost

A rule of thumb: one task per session. When you finish, commit, clear and start the next one with a short prompt that points at the files that matter.

A day with the cheat sheet

Here is how the pieces fit in an ordinary day, so the tables above have a story behind them.

Morning. claude -c picks up yesterday's session, or claude starts a new one. Run /context to see what loaded.
First task. Switch to plan mode, ask for a read-only summary of the area you will touch, then leave plan mode and make the change.
Midday. Commit after each green test run. If a session sprawls, /compact it or /clear and restart from your notes.
A side question. Use a subagent or a second terminal, so the main thread stays clean.
Reviewing. Pipe a diff into claude -p for a quick second opinion, and read the findings yourself.
End of day. Update CLAUDE.md with anything you had to tell the agent twice, and commit it.

Keep this page open for the first week, and by the second you will need only the short list under Quick facts.

Common problems

Symptom	Try
Config does not seem to apply	`/doctor`, and `/context`; the docs have a debug-your-config page
A skill or command is missing	`/reload-skills`, or check the folder name for odd characters with `claude --debug`
An MCP server shows as pending	Project servers need your approval; `claude mcp reset-project-choices` resets choices
MCP server is slow to start	Raise the `MCP_TIMEOUT` environment variable
Output looks garbled	`Ctrl+L` redraws
Alt shortcuts do nothing on a Mac	Enable Option as Meta in your terminal
Context fills fast	`/context`, then `/compact` or `/clear`

Our take

We compiled this from the docs and did not time any workflow. Treat it as a map, not a benchmark.

If you only memorise four things, make them Esc to stop, Shift+Tab to change mode, /compact and /clear to manage context, and claude -c to resume. Together they cover most of the friction of a long session.

The -p flag with --output-format json, --max-turns and --max-budget-usd is the part most people skip, and it turns Claude Code from a chat into a scriptable tool with a spending cap. If you run it in CI, start there.

Command lists change quickly. The docs mark several commands as removed and add new ones every few releases. Run claude --help, /help and claude update before you trust a screenshot, including ours. A cheat sheet is only as current as the date on it, and ours says 31 October 2026.

Frequently asked questions

How do I run Claude Code once and exit? claude -p "your question". Pipe content in with cat file | claude -p "explain".

How do I continue my last conversation? claude -c, or claude -r with a session ID or name.

What does Shift+Tab do? It cycles permission modes.

How do I stop Claude mid-response? Press Esc. Press it twice to clear a draft or open rewind.

Where do project settings and MCP servers live? .claude/settings.json and .mcp.json.

*Product names are used descriptively. We are not affiliated with Anthropic.*

Questions

How do I run Claude Code once and exit?

Use claude -p 'your question'. The -p flag prints the response and exits, and you can pipe content in with cat file | claude -p 'explain'.

How do I continue my last conversation?

Run claude -c for the most recent conversation in the current directory, or claude -r with a session ID or name to pick one.

What does Shift+Tab do?

It cycles permission modes: default, acceptEdits, plan, and when available bypassPermissions and auto.

How do I stop Claude mid-response?

Press Esc. Claude keeps the work done so far so you can redirect. Press Esc twice to clear a draft or open rewind.

Where do project settings and MCP servers live?

Project settings are in .claude/settings.json, and team-shared MCP servers are in .mcp.json at the project root.

Loading poll...

Claude Code Cheat Sheet

Quick facts

Start and run

Flags you will use

Print mode only (`-p`)

Keyboard shortcuts

General

Editing the prompt

Multiline input

Slash commands

The .claude folder

MCP in one screen

Skills, subagents and plugins in one screen

Recipes

A starter settings.json

Which permission mode when

Cost and context habits

A day with the cheat sheet

Common problems

Our take

Frequently asked questions

Questions

Sources

Part of the Wbcom Designs family

Claude Code Cheat Sheet

Quick facts

Start and run

Flags you will use

Print mode only (-p)

Keyboard shortcuts

General

Editing the prompt

Multiline input

Slash commands

The .claude folder

MCP in one screen

Skills, subagents and plugins in one screen

Recipes

A starter settings.json

Which permission mode when

Cost and context habits

A day with the cheat sheet

Common problems

Our take

Frequently asked questions

Questions

Sources

Part of the Wbcom Designs family

Print mode only (`-p`)