Claude Code — AI Coding Agent for the Terminal

Claude Code is Anthropic's official agentic coding tool: it reads, writes and runs code right in your terminal, driven by the claude command. By default it opens an interactive session in the current directory; with -p/--print it turns into a scriptable, pipe-friendly one-shot tool. Beyond the CLI it also runs as a desktop and web app and as an IDE extension, and it taps the latest Claude models — currently Opus 5.5, Sonnet 5.5, Fable 5.1 and Haiku 5.5. This cheat sheet collects the flags, subcommands and slash commands you reach for every day, from session management, background and cloud sessions and permission modes to MCP servers, subagents, plugins and git worktrees. As of October 2026, Claude Code 2.1.294 — Claude Code moves fast, so a look at claude --help and /help pays off after every update.

Basic Usage

claude — Start an interactive Claude Code session in the current directory.

claude

claude "<prompt>" — Start an interactive session with an initial prompt.

claude "Explain the architecture of this project"

claude --version — Show the installed Claude Code version.

claude --version

claude --help — Show all available options and commands.

claude --help

claude update — Check for updates and install the latest version if available. Alias: claude upgrade

claude update

claude doctor — Check the health of the installation without starting a session (reads settings files in the current directory without a trust prompt). For a full checkup that can also fix issues, run /doctor in a session.

claude doctor

claude install [target] — Install a specific Claude Code native build (stable, latest, or a version number). --force reinstalls even if it is already installed.

claude install stable

claude purge [path] — Delete all local Claude Code state for a project: transcripts, task lists, debug logs, file-edit history, prompt history lines and its entry in ~/.claude.json. Without a path an interactive list opens; --dry-run previews, -i confirms each item, -y skips the confirmation, --all purges every project (cannot be combined with a path). Renamed from claude project purge in v2.1.288 (the old name still works and prints a notice).

claude purge --dry-run ~/projects/old-app

claude gateway --config <path> — Run the enterprise auth/telemetry gateway from a YAML config.

claude gateway --config ./gateway.yaml

claude ultrareview [target] — Run a cloud-hosted multi-agent code review of the current branch (or a PR number / base branch) and print the findings. --post posts them to the PR as one comment, --json prints the raw payload, --timeout sets the wait limit in minutes (default 45). In a session, use /code-review ultra.

claude ultrareview 42

claude import [source] — Import configuration from another AI coding agent — codex, gemini or cursor: instruction files, MCP servers, commands, subagents and skills. --dry-run shows what would be imported without writing anything.

claude import codex --dry-run

Session Management

claude -c — Continue the most recent conversation in the current directory. Long: --continue

claude -c

claude -r — Open an interactive picker to resume a previous conversation, optionally filtered by a search term. Long: --resume

claude -r

claude -r <session-id|name> — Resume a specific conversation directly by its session ID or name.

claude -r "auth-refactor" "Finish this PR"

claude -n <name> — Set a display name for this session (shown in the prompt box, the /resume picker and the terminal title). Resume it later with claude --resume <name>. Long: --name

claude -n "refactor auth module"

claude --fork-session — When resuming, create a new session ID instead of continuing the original.

claude -c --fork-session

claude --session-id <uuid> — Use a specific UUID as the session ID.

claude --session-id 550e8400-e29b-41d4-a716-446655440000

claude --from-pr <number> — Resume a session linked to a pull request by number or URL; without a value, a picker opens.

claude --from-pr 42

Background Sessions

claude --bg "<prompt>" — Start the session in the background and return immediately. Prints the short ID that claude attach, logs, stop and rm take; claude agents lists all sessions. With --resume <session-id>, continues that session in the background. Long: --background

claude --bg "Run the full test suite and summarize failures"

claude agents — Open agent view to monitor and dispatch parallel background sessions. In the view, n:<text> filters by session name or task (v2.1.287 or later); Ctrl+F finds a session by name, Enter opens the best match, and Alt+↑/↓ jumps between groups (v2.1.288 or later). --cwd <path> shows only sessions started under that directory; --model, --effort and --permission-mode set defaults for the sessions you dispatch.

claude agents --cwd ~/projects/app

claude agents --json — Print active sessions (interactive and background) as a JSON array for scripting; --all also includes completed background sessions.

claude agents --json --all

claude attach <id|name> — Open a background session in this terminal. Takes the short ID that claude --bg prints and claude agents lists, or part of the session name — quote names with spaces (v2.1.290 or later). ← returns to agent view, Ctrl+Z drops back to the shell; the session keeps running.

claude attach "auth refactor"

claude logs <id|name> — Print a background session's recent terminal output. Part of the session name works in place of the ID — quote names with spaces (v2.1.290 or later).

claude logs auth

claude stop <id> — Stop a background session. Its conversation is kept: claude attach <id> opens it again, and claude --resume works once it is stopped. Alias: claude kill

claude stop 7c5dcf5d

claude rm <id> — Delete a background session, and its worktree when that is safe. Works on sessions that have already exited.

claude rm 7c5dcf5d

claude respawn <id> — Restart a background session with its conversation intact so it runs the current Claude Code version; --all restarts every running session.

claude respawn --all

claude --bg --exec '<command>' — Run a shell command as a PTY-backed background job instead of starting a Claude session.

claude --bg --exec 'pytest -x'

claude daemon status — Print the background-session supervisor's state, version, socket directory and worker count for diagnostics.

claude daemon status

claude daemon stop --any — Stop the background-session supervisor and the sessions it hosts, to recover from a stuck supervisor; --keep-workers leaves the sessions running.

claude daemon stop --any --keep-workers

Non-Interactive Mode (--print)

claude -p "<prompt>" — Print the response and exit. Non-interactive for scripting and pipes.

claude -p "What does this function do?" < src/utils.js

echo "<prompt>" | claude -p — Pipe a prompt into Claude and get a single printed response.

echo "Summarize this" | claude -p

cat <file> | claude -p "<prompt>" — Pipe file contents as context into a non-interactive prompt.

cat app.py | claude -p "Find security issues"

claude -p --output-format json "<prompt>" — Output the response as a single JSON object. Formats: text, json, stream-json.

claude -p --output-format json "List 3 refactoring tips"

claude -p --output-format stream-json --verbose "<prompt>" — Stream the response as newline-delimited JSON events in real time. With -p, stream-json also requires --verbose.

claude -p --output-format stream-json --verbose "Explain async/await"

claude -p --json-schema '<schema>' "<prompt>" — Validate output against a JSON Schema for structured, machine-readable responses.

claude -p --json-schema '{"type":"object","properties":{"name":{"type":"string"}}}' "Extract info"

claude -p --max-budget-usd <amount> "<prompt>" — Set a maximum spend limit in USD for this call.

claude -p --max-budget-usd 0.10 "Summarize these logs"

claude -p --max-turns <n> "<prompt>" — Limit the number of agentic turns; exits with an error when the limit is reached. No limit by default.

claude -p --max-turns 5 "Fix the failing test"

claude -p --permission-prompts none "<prompt>" — Deny every action that would need a permission prompt instead of waiting for an answer; the permission mode still decides everything else. The default host sends prompts to the SDK host or --permission-prompt-tool.

claude -p --permission-prompts none "Run the linter and fix what it finds"

claude -p --permission-prompt-tool <mcp-tool> "<prompt>" — Let an MCP tool answer permission prompts in non-interactive mode.

claude -p --permission-prompt-tool mcp__approver__check "Deploy to staging"

claude -p --no-session-persistence "<prompt>" — Don't save the session to disk (cannot be resumed).

claude -p --no-session-persistence "Quick question"

claude --fallback-model <models> — Automatically fall back to the specified model when the default is overloaded or unavailable. Takes a comma-separated list tried in order; the primary model is retried at the start of each user turn. Works in interactive sessions too; the fallbackModel setting makes it permanent.

claude -p --fallback-model sonnet,haiku "Generate unit tests"

claude -p --input-format stream-json --output-format stream-json --verbose — Bidirectional streaming JSON for integration with custom tools or orchestrators (stream-json output needs --verbose).

claude -p --input-format stream-json --output-format stream-json --verbose

claude -p --include-partial-messages ... — Emit partial message chunks as they arrive (requires --output-format stream-json and --verbose).

claude -p --output-format stream-json --verbose --include-partial-messages "Explain this"

claude -p --include-hook-events ... — Include all hook lifecycle events in the output stream (requires --output-format stream-json and --verbose).

claude -p --output-format stream-json --verbose --include-hook-events "Run the build"

claude -p --replay-user-messages ... — Re-emit user messages from stdin back on stdout for acknowledgment (stream-json in and out, plus --verbose).

claude -p --input-format stream-json --output-format stream-json --verbose --replay-user-messages

claude -p --prompt-suggestions "<prompt>" — Emit a prompt_suggestion message after each turn predicting the next user prompt (requires --output-format stream-json and --verbose; very short conversations may produce none). --prompt-suggestions=false switches it off.

claude -p --output-format stream-json --verbose --prompt-suggestions "Refactor this module"

claude -p --forward-subagent-text ... — Forward subagent text and thinking blocks as their own messages with parent_tool_use_id set (requires --output-format stream-json and --verbose).

claude -p --output-format stream-json --verbose --forward-subagent-text "Audit the codebase"

Model & Effort

claude --model opus — Use the latest Opus — Opus 5.5 on the Anthropic API (requires v2.1.280 or later). The aliases opus, sonnet, haiku and fable always point to the recommended version of their line; on other providers (Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry) some resolve to older models, e.g. sonnet to Sonnet 4.5 or 4.6 and haiku to Haiku 4.5.

claude --model opus

claude --model fable — Use Claude Fable 5.1 (v2.1.257 or later), the most capable model, for the hardest and longest-running tasks. Depending on your plan, Fable usage can bill to usage credits.

claude --model fable

claude --model sonnet — Use the latest Sonnet — Sonnet 5.5 on the Anthropic API (v2.1.284 or later), for everyday coding.

claude --model sonnet

claude --model haiku — Use the latest Haiku — Haiku 5.5 on the Anthropic API (v2.1.293 or later), the fastest and most cost-efficient model, for simple tasks; 1M context, default effort medium. Other providers still resolve haiku to Haiku 4.5.

claude --model haiku

claude --model best — Use the most capable model available to you: Fable where you have access, otherwise the same model as opus.

claude --model best

claude --model opusplan — Use Opus while in plan mode, then switch to Sonnet for execution.

claude --model opusplan

claude --model 'opus[1m]' — Request the 1M-token context window explicitly (sonnet[1m] likewise). Only needed for models that reach 1M only through a [1m] variant, such as Opus 4.6 or Sonnet 4.6; Opus 4.7+, Sonnet 5+, Haiku 5.5 and Fable already run with 1M natively on the Anthropic API and behind gateways (Opus 4.7+ and Fable also on Bedrock, Vertex and Foundry since v2.1.287).

claude --model 'opus[1m]'

claude --model <full-model-id> — Use a specific model by its full ID. Current: claude-opus-5-5, claude-sonnet-5-5, claude-fable-5-1, claude-haiku-5-5; older ones such as claude-opus-5, claude-fable-5 or claude-haiku-4-5 (pinned snapshot claude-haiku-4-5-20251001) remain selectable.

claude --model claude-opus-5-5

claude --effort <level> — Set the effort level for this session: low, medium, high, xhigh, max — available levels depend on the model. ultracode starts at xhigh with ultracode (dynamic workflows) on. Not saved as a default.

claude --effort high "Perform a full security audit"

claude --advisor <model> — Enable the server-side advisor tool for this session, which consults a second model (fable, opus, sonnet or a full ID such as claude-haiku-5-5) at key moments. Haiku 5.5 can advise from v2.1.293; Haiku 4.5 can't.

claude --advisor opus

claude --autocompact <auto|tokens> — Set the auto-compact window for this session — auto, or 100k–1M tokens — without changing your saved settings.

claude --autocompact 500k

Permission Modes & Tool Access

claude --permission-mode manual — Manual mode: asks before every action except reads. Its config value is default, which the CLI also accepts. Without a flag or defaultMode setting, non-interactive runs (-p) and the Agent SDK start in this mode.

claude --permission-mode manual

claude --permission-mode acceptEdits — Auto-accept file edits and common filesystem commands (mkdir, touch, mv, cp, rm …) inside the working directories; protected paths and anything outside still prompt.

claude --permission-mode acceptEdits

claude --permission-mode plan — Plan mode: propose changes but don't execute until approved.

claude --permission-mode plan

claude --permission-mode auto — Auto mode: a classifier reviews each action and approves safe ones without prompting. Since v2.1.284, new terminal and VS Code sessions start in auto mode on every plan and provider unless a flag or defaultMode says otherwise; where auto mode isn't available, they start in Manual. Shift+Tab cycles through the modes.

claude --permission-mode auto

claude --permission-mode dontAsk — Never prompt: anything that would need approval is denied automatically. Reads, read-only commands and calls matching your allow rules still run — made for CI with an exact allowlist.

claude --permission-mode dontAsk

claude --permission-mode bypassPermissions — Skip permission prompts entirely; deny rules and a few safeguards (ask rules, removal of critical paths) still apply. Only for isolated sandboxes.

claude --permission-mode bypassPermissions

claude --dangerously-skip-permissions — Same as --permission-mode bypassPermissions: skips permission prompts, while deny rules and a few safeguards still apply. Only for sandboxes with no internet access.

claude --dangerously-skip-permissions

claude --allow-dangerously-skip-permissions — Make 'bypass permissions' available as an opt-in mode without enabling it by default.

claude --allow-dangerously-skip-permissions

claude --allowed-tools "<tools>" — Tools that run without a permission prompt, with patterns such as Bash(git *). To limit which tools exist at all, use --tools.

claude --allowed-tools "Bash(git *) Edit Read"

claude --disallowed-tools "<tools>" — Deny rules: a bare tool name removes the tool entirely, a scoped rule such as Bash(rm *) only denies matching calls.

claude --disallowed-tools "Bash(rm *) WebFetch"

claude --tools "<tools>" — Restrict which built-in tools exist in the session: "" disables all, "default" restores the default set (which on macOS, Linux and WSL leaves out Glob and Grep unless you name them). MCP tools are not affected.

claude --tools "Read,Glob,Grep"

claude --add-dir <path> — Grant tool access to additional directories beyond the current working directory.

claude --add-dir ~/shared-libs --add-dir /etc/nginx

claude --restricted — Restricted mode for shared machines and evaluation harnesses: removes the built-in tools that run commands or code and WebFetch (unless --tools names them), ignores user, project and local settings files, confines the file tools to the working directories and refuses bypassPermissions.

claude --restricted -p "Review this diff"

System Prompts & Agents

claude --system-prompt "<prompt>" — Override the default system prompt for this session.

claude --system-prompt "Reply only in German."

claude --append-system-prompt "<prompt>" — Append instructions to the default system prompt without replacing it.

claude --append-system-prompt "Always use Bootstrap 5."

claude --system-prompt-file <path> — Load the system prompt override from a file instead of an inline string.

claude --system-prompt-file ./prompts/reviewer.md

claude --append-system-prompt-file <path> — Append a system prompt loaded from a file to the default system prompt.

claude --append-system-prompt-file ./prompts/house-rules.md

claude --brief — Enable the SendUserMessage tool so an agent can send messages back to the user.

claude --brief

claude --agent <agent-name> — Run the session as the named agent — a subagent definition from .claude/agents/, a plugin or --agents. Overrides the agent setting.

claude --agent reviewer

claude --agents '<json>' — Define inline agents as JSON for the current session; with --print, a path to a JSON file also works.

claude --agents '{"tester":{"description":"Writes tests","prompt":"Write unit tests only"}}'

claude --bare — Minimal mode: skip hooks from settings and plugins, LSP, plugin sync, attribution, auto-memory, background prefetches, keychain reads and CLAUDE.md discovery; sets CLAUDE_CODE_SIMPLE=1. Since v2.1.286 it connects only the MCP servers passed on the command line (--mcp-config), sends the model no system reminders and starts no background tasks; a shell command that hits its timeout stops instead of moving to the background. Anthropic auth is strictly ANTHROPIC_API_KEY or apiKeyHelper — OAuth and the keychain are never read; Bedrock, Vertex and Foundry use their own credentials. Provide context explicitly (--system-prompt[-file], --append-system-prompt[-file], --add-dir, --mcp-config, --settings, --agents, --plugin-dir).

claude --bare -p "Quick question" --mcp-config ./mcp.json

claude --init-only — Run the Setup and SessionStart hooks, then exit without starting a conversation — useful for CI and container images. --init and --maintenance run Setup hooks before a -p session.

claude --init-only

claude --system-prompt-snapshot off — Rebuild the system prompt on every request instead of reusing the one recorded at the start of the conversation (default on) — useful while iterating on prompt text.

claude --append-system-prompt "Draft rules" --system-prompt-snapshot off

claude -p --append-subagent-system-prompt "<text>" — Append text to every subagent's system prompt (print mode only; --append-subagent-system-prompt-file reads it from a file).

claude -p --append-subagent-system-prompt "Answer in German." "Audit the repo"

Interactive Slash Commands: Session

/help — Show help and available commands.

/help

/clear — Start a new conversation with empty context; pass a name to label the previous one in the /resume picker. Aliases: /reset, /new

/clear

/compact [instructions] — Compact conversation to save context, with optional focus instructions.

/compact focus on the auth module changes

/context [all] — Visualize current context usage as a colored grid; all expands the per-item breakdown in fullscreen mode.

/context

/resume [session] — Resume a conversation by ID or name. Alias: /continue

/resume

/branch [name] — Branch the conversation at this point and switch into the branch; the original stays available through /resume.

/branch experiment-1

/fork [prompt] — Copy the conversation into a new background session and keep working here; with a prompt, the copy starts on it right away.

/fork try the alternative approach

/subtask <task> — Hand a side task to a forked subagent that inherits the full conversation; its result comes back into this conversation when it finishes.

/subtask write tests for the new parser

/background [prompt] — Detach the current session to run as a background agent and free this terminal. Alias: /bg

/background finish the test suite

/goal [condition|clear] — Set a goal: Claude keeps working until the condition is met. Useful for long autonomous runs.

/goal all tests pass and the linter is clean

/cd <path> — Move the running session to a different working directory.

/cd ~/projects/other-repo

/teleport — Pull a cloud session into this terminal, including its branch and conversation. Alias: /tp; requires a claude.ai subscription.

/teleport

/rename [name] — Rename the current session; without a name, one is generated from the conversation.

/rename auth-refactor

/export [filename] — Export the current conversation as plain text.

/export conversation.txt

/recap — Generate a one-line summary of the current session on demand.

/recap

/stop — Stop the background session you are attached to; the transcript and any worktree are kept. /exit only detaches.

/stop

/exit — Exit the CLI; in an attached background session it detaches and the session keeps running. Alias: /quit

/exit

Interactive Slash Commands: Model & Tools

/model [model] — Switch the model and save it as your default for new sessions; in the picker, press s to switch for this session only.

/model opus

/effort [level|auto|status|ultracode] — Set the effort level: low, medium, high or xhigh (saved as the default for this model), max (this session only) or auto; status prints it. ultracode [on|off] switches ultracode — dynamic workflow orchestration — on or off for the session at the current level.

/effort ultracode on

/advisor [model|off] — Enable the advisor tool, which consults a second model (fable, opus, sonnet or a full ID such as claude-haiku-5-5) for guidance at key moments; off disables it. Experimental, Anthropic API only; fable requires Fable access.

/advisor opus

/fast [on|off] — Toggle fast mode: the same Opus model at up to 2.5x output speed for a higher price per token, with no downgrade to a smaller model. Supported on Opus 5.5 (fast mode's default since v2.1.280), Opus 5 and Opus 4.8; on subscription plans it bills to usage credits only. Toggled mid-turn, it takes effect from the next turn.

/fast on

/usage — Show session cost, plan usage limits and activity stats. Aliases: /cost, /stats (opens on the Stats tab)

/usage

/permissions — Manage allow, ask and deny rules, working directories and the auto mode rules. Alias: /allowed-tools

/permissions

/tasks — View and manage background work in the current session, including finished subagents. Alias: /bashes

/tasks

/autocompact [auto|<tokens>] — Set how full the context window gets before automatic compaction, e.g. 500k, or auto to return to the window tuned for your model. Saved to user settings per model (under modelSettings), so each model keeps its own window when you switch (v2.1.288 or later).

/autocompact 500k

/sandbox — Toggle sandbox mode (supported platforms only).

/sandbox

Interactive Slash Commands: Code & Git

/diff — Open interactive diff viewer showing uncommitted changes.

/diff

/rewind — Rewind the conversation and/or code to a previous point, or summarize from a selected message. Aliases: /checkpoint, /undo

/rewind

/code-review [level] [--fix] [--comment] [target] — Review the current diff — or a PR number, branch or path — for correctness bugs. Levels: low, medium, high, xhigh, max, or ultra (cloud-hosted multi-agent review that needs a claude.ai account; 3 free runs on Pro/Max, then usage credits; /ultrareview is an alias). --fix applies the findings, --comment posts them on the GitHub PR or GitLab merge request; --max-findings <n>|all reports more or fewer findings than the usual limit and is remembered until --max-findings default (v2.1.288 or later).

/code-review high --max-findings all

/review [level] [--fix] [--comment] [target] — Alias of /code-review, with the same levels, flags and targets.

/review 42

/security-review — Analyze the diff between your branch and origin's default branch for security vulnerabilities such as injection, auth issues and data exposure.

/security-review

/copy [N] — Copy the last (or Nth-latest) response to the clipboard; with code blocks, a picker opens where w writes the selection to a file.

/copy

/btw [question] — Ask a quick side question without adding it to the conversation; without a question, browse your earlier side answers.

/btw what's the syntax for a ternary in PHP?

Interactive Slash Commands: Configuration

/config [key=value ...] — Open the settings (theme, model, output style, editor mode and more) or set values directly, e.g. /config theme=dark. The Vim editing mode lives here now; /vim was removed. Alias: /settings

/config model=sonnet

/status — Open the settings on the Status tab: version, model, account and connectivity.

/status

/theme — Change the color theme: auto (follows the terminal background), light and dark variants, colorblind-accessible and ANSI themes, and custom themes.

/theme

/color [color|default] — Set the prompt bar color: red, blue, green, yellow, purple, orange, pink or cyan; default resets it, no argument picks a random color.

/color blue

/voice [hold|tap|off] — Toggle voice dictation or pick a mode: hold, tap or off. Requires a Claude.ai account.

/voice tap

/keybindings — Open or create keybindings configuration file.

/keybindings

/terminal-setup — Set up the newline key for your terminal: Shift+Enter in VS Code, Cursor, Alacritty or Zed; Option+Enter in Apple Terminal; clipboard access in iTerm2.

/terminal-setup

/tui [default|fullscreen] — Set the terminal UI renderer and relaunch into it with the conversation intact: fullscreen (flicker-free alt-screen renderer, saved as the tui setting — CLAUDE_CODE_NO_FLICKER=1 does the same for a single launch) or default. Without an argument, prints the active renderer.

/tui fullscreen

/statusline — Configure Claude Code's status line (or auto-configure from shell prompt).

/statusline

/focus — Toggle the focus view: only your last prompt, a one-line tool summary and the final response (fullscreen rendering only).

/focus

/debug [description] — Enable debug logging for the current session and troubleshoot issues.

/debug mcp server keeps disconnecting

/heapdump — Write a JavaScript heap snapshot and a memory breakdown to ~/Desktop (or your home directory) for diagnosing memory usage. Hidden from the menu — type it in full; share only the -diagnostics.json file.

/heapdump

/memory — Edit CLAUDE.md, enable/disable auto-memory, view auto-memory entries.

/memory

/output-style [style] — List output styles or switch to one.

/output-style concise

/init — Initialize the project with a CLAUDE.md guide; CLAUDE_CODE_NEW_INIT=1 starts an interactive flow that also covers skills, hooks and personal memory files.

/init

/hooks — View hook configurations for tool events.

/hooks

/reload-skills — Re-scan skill and command directories so skills added or changed on disk become available without a restart.

/reload-skills

/add-dir <path> — Add a new working directory to the current session.

/add-dir ~/other-project

Interactive Slash Commands: Integrations

/mcp [reconnect <server>|enable|disable] — Manage MCP server connections and OAuth; reconnect <server> or enable/disable with a server name or all act without opening the dialog.

/mcp reconnect sentry

/ide — Manage IDE integrations (VS Code, JetBrains) and show status.

/ide

/chrome — Configure Claude in Chrome integration.

/chrome

/plugin [subcommand] — Manage Claude Code plugins: the menu without an argument, or list, install, enable, disable directly.

/plugin

/reload-plugins [--force] — Reload all active plugins to apply pending changes without restarting; if the reload would change MCP tools and invalidate the prompt cache, it warns and skips unless you pass --force.

/reload-plugins

/skills — List available skills; filter by typing, sort by token cost with t, and toggle what Claude and the / menu see.

/skills

/skill-doctor — Show what each skill costs in context and how often it is used, to find skills worth turning off.

/skill-doctor

/agents — Since v2.1.198 only a reminder: ask Claude to create or manage subagents, or edit .claude/agents/ or ~/.claude/agents/ directly.

/agents

/plan [description] — Enter plan mode directly from the prompt.

/plan refactor the authentication module

/schedule [description] — Create, update, list or run routines, which execute in the cloud; requires a claude.ai subscription. Alias: /routines

/schedule

/import [codex|gemini|cursor] — Bring instruction files, MCP servers, commands, subagents and skills over from OpenAI Codex, Gemini CLI or Cursor.

/import cursor

/remote-env — Choose the default cloud environment for cloud sessions you start from the CLI.

/remote-env

/web-setup — Connect your GitHub account for cloud sessions using your local gh credentials.

/web-setup

/workflows — Open the workflow progress view to watch, pause, resume or save running and completed workflows.

/workflows

/artifacts — List the artifacts you own or that are shared with you, then attach one to the session, open it in the browser or copy its link (where artifacts are available).

/artifacts

/list-agents — List the subagents, teammates and other sessions Claude can message, with the name to use for each (cross-session messaging only). Alias: /peers

/list-agents

Interactive Slash Commands: Account & Info

/login — Sign in to your Anthropic account.

/login

/logout — Sign out from your Anthropic account.

/logout

/doctor — Skill: a setup checkup that diagnoses issues and can fix them — installation, settings files, unused skills, MCP servers and plugins, slow hooks, updates; asks before changing anything. /doctor prompt-audit checks CLAUDE.md and skills for outdated or conflicting instructions. Alias: /checkup

/doctor

/insights — Generate an HTML report analyzing your recent sessions on this machine: projects, usage patterns, what goes wrong and features to try.

/insights

/release-notes — Browse the changelog in an interactive version picker.

/release-notes

/feedback [report] — Send feedback or report a bug: opens the same consent dialog as /bug, where you choose how much session history to include. Without an argument, it opens the queue of feedback Claude has drafted, if there is any.

/feedback

/bug [report] — Report a bug or share your conversation; you choose how much session history to include and confirm before anything is sent. Without Anthropic credentials the report is written to ~/.claude/feedback-bundles/. Alias: /share

/bug the diff viewer freezes on large files

/desktop — Continue the current session in the Claude Code Desktop app (macOS or x64 Windows, Claude subscription). Alias: /app

/desktop

/mobile — Show QR code to download Claude mobile app. Aliases: /ios, /android

/mobile

/remote-control — Make the session available for Remote Control from claude.ai (claude.ai subscription; not with API keys). Alias: /rc

/remote-control

/install-github-app — Install the Claude GitHub App for a repository, optionally with GitHub Actions workflows and secrets (github.com only).

/install-github-app

/usage-credits — Configure usage credits — or request them from your admin — when you hit a limit (formerly /extra-usage, which still works).

/usage-credits

/rate-limit-options — Show ways to keep working when a usage limit blocks a request: wait and continue automatically, add usage credits, or upgrade your plan (claude.ai subscription).

/rate-limit-options

/team-onboarding — Generate an onboarding guide for teammates from your Claude Code usage over the past 30 days.

/team-onboarding

/powerup — Discover Claude Code features through quick interactive lessons with animated demos.

/powerup

/upgrade — Open the upgrade page to switch to a higher plan.

/upgrade

/privacy-settings — View and update your privacy settings (Pro and Max).

/privacy-settings

/setup-bedrock — Interactive wizard for Amazon Bedrock authentication, region and model pins; hidden until CLAUDE_CODE_USE_BEDROCK=1 is set (/setup-vertex likewise for Google Cloud with CLAUDE_CODE_USE_VERTEX=1).

/setup-bedrock

Bundled Skills & Workflows

/batch <instruction> — Orchestrate large-scale changes across the codebase in parallel.

/batch replace all deprecated API calls with the v2 equivalent

/loop [interval] [prompt] — Run a prompt repeatedly while the session stays open. Without an interval the model paces itself; without a prompt it runs the built-in maintenance prompt or your loop.md. Alias: /proactive

/loop 5m check whether the deploy finished

/verify — Confirm a change does what it should by building, running and observing your project's app instead of relying on tests alone (runs only when you invoke it).

/verify

/run — Launch and drive your project's app to see a change working, not only passing tests.

/run

/run-skill-generator — Teach /run and /verify how to build, launch and drive your project's app by writing a per-project skill.

/run-skill-generator

/claude-in-chrome [task] — Have Claude carry out a task in your browser — test a page, fill a form, read console logs — through Claude in Chrome.

/claude-in-chrome check the signup form for console errors

/simplify [target] — Four parallel review agents look for reuse of existing helpers, simplification, efficiency and the right level of abstraction, then apply the fixes. No bug hunting — that is /code-review.

/simplify src/auth/

/update-config [request] — Describe a settings change — allow a command, set an environment variable, add a hook — and Claude edits the matching settings.json.

/update-config allow npm test without asking

/plugin-authoring [request] — Load the reference Claude works from to write a mod: a live pane, band, status line, toast or hook inside Claude Code, built as a plugin of function hooks that hot-reloads in the running session. Claude also loads it on its own when you ask for a mod, and since v2.1.290 gives you the one command others run to install it. Requires v2.1.287 or later; it comes from a built-in plugin you can turn off in /plugin.

/plugin-authoring a mod that shows the current git branch above the prompt

/claude-api [subcommand] — Load Claude API and Managed Agents reference material for your project's language; subcommands such as migrate or cost-optimize focus it. Since v2.1.290, managed-agents-onboard <url> sets up the Managed Agents pattern a page describes as ant apply files, and managed-agents-onboard <quickstart-name> builds a Console quickstart such as deep-researcher with the ant CLI.

/claude-api managed-agents-onboard deep-researcher

/design [brief] — Draft UI mockups, screen flows or landing pages as artboards on a Claude Design canvas (needs a session where artifacts are available).

/design a settings screen for a mobile banking app

/slides [brief] — Make a new presentation as a Claude Slides artifact filled from your brief (needs a session where artifacts are available).

/slides a quarterly review of the platform team

/fewer-permission-prompts — Scan your transcripts for common read-only Bash and MCP calls and add a prioritized allowlist to the project's .claude/settings.json.

/fewer-permission-prompts

/dataviz [request] — Design guidance for charts, graphs and dashboards.

/dataviz a bar chart comparing quarterly revenue

/deep-research <question> — Workflow: fan out web searches, cross-check sources and write a cited report. Requires the WebSearch tool; on Pro, turn on dynamic workflows in /config first.

/deep-research how do European accessibility laws differ from WCAG?

/autofix-pr [prompt] — Spawn a cloud session that watches the current branch's PR and pushes fixes when CI fails or reviewers leave comments (requires the gh CLI and access to cloud sessions).

/autofix-pr

MCP Servers

claude mcp list — List all configured MCP (Model Context Protocol) servers.

claude mcp list

claude mcp add <name> -- <command> [args...] — Add a stdio MCP server as a subprocess.

claude mcp add my-server -- npx my-mcp-server --port 3000

claude mcp add --transport http <name> <url> — Add an HTTP/SSE MCP server by URL.

claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

claude mcp add <name> -e <KEY>=<val> -- <command> — Add a stdio MCP server with environment variables. Put -e after the server name: it takes several pairs and would read a name that follows it as another pair.

claude mcp add my-server -e API_KEY=xxx -- npx -y my-mcp-server

claude mcp add -s <scope> <name> ... — Choose where the server is stored: local (default, only you in this project), project (shared through .mcp.json) or user (all your projects). Long: --scope

claude mcp add -s user --transport http sentry https://mcp.sentry.dev/mcp

claude mcp add --transport http <name> <url> --header "Authorization: Bearer <token>" — Add an HTTP MCP server with a custom auth header.

claude mcp add --transport http api https://api.example.com/mcp --header "Authorization: Bearer xxx"

claude mcp add-json <name> '<json>' — Add an MCP server (stdio, SSE, HTTP or WebSocket) from a full JSON config string.

claude mcp add-json myserver '{"type":"stdio","command":"npx","args":["my-server"]}'

claude mcp get <name> — Show full configuration details of a specific MCP server.

claude mcp get sentry

claude mcp remove <name> — Remove a configured MCP server.

claude mcp remove my-server

claude mcp login <name> — Run a configured server's OAuth flow without the /mcp panel (HTTP, SSE and claude.ai connectors); --no-browser prints the URL instead, for SSH.

claude mcp login sentry --no-browser

claude mcp logout <name> — Clear stored OAuth credentials for an MCP server.

claude mcp logout sentry

claude mcp add-from-claude-desktop — Import MCP servers from the Claude Desktop app (Mac and WSL only).

claude mcp add-from-claude-desktop

claude mcp serve — Start Claude Code itself as an MCP server (for integration with other tools).

claude mcp serve

claude --mcp-config <path> — Load additional MCP servers from a JSON config file for this session.

claude --mcp-config ~/.config/mcp-servers.json

claude --strict-mcp-config --mcp-config <path> — Only use MCP servers from the specified config, ignore all others.

claude --strict-mcp-config --mcp-config config.json

claude mcp reset-project-choices — Reset all approved/rejected project-scoped MCP server choices.

claude mcp reset-project-choices

Authentication

claude auth login — Sign in to your Anthropic account (opens the browser). --console bills through the Anthropic Console instead of a Claude subscription, --sso forces SSO, --email pre-fills the address.

claude auth login

claude auth logout — Log out from the current Anthropic account.

claude auth logout

claude auth status — Show the current authentication status and active account — as JSON by default, or readable with --text.

claude auth status --text

claude setup-token — Set up a long-lived authentication token (requires Claude subscription).

claude setup-token

Worktrees, IDE & Cloud

claude -w — Create a new git worktree for this session (isolated branch). Long: --worktree

claude -w

claude -w <name> — Create a named git worktree.

claude -w feature-refactor

claude -w --tmux — Create a worktree and open it in a new tmux session (iTerm2 native panes when available; --tmux=classic forces traditional tmux).

claude -w --tmux

claude --ide — Automatically connect to the IDE on startup if exactly one valid IDE is available.

claude --ide

claude --cloud "<task>" — Create a cloud session for the task; with a session ID or claude.ai/code URL (and -p), queue a message into an existing one instead. --remote is a deprecated alias.

claude --cloud "Fix the login bug"

claude --environment <id> — Create a new cloud session that runs on a self-hosted environment (IDs start with ccpool_).

claude -p "Fix the login bug" --environment ccpool_abc123

claude --teleport [session] — Resume a cloud session in your local terminal.

claude --teleport

claude --desktop — Open in the Claude Desktop app instead of the terminal, with --continue or --resume <id> to pick the session (v2.1.285+, macOS and x64 Windows only, Claude account sign-in required, no prompt argument).

claude --desktop --continue

claude --remote-control [name] — Start an interactive session with Remote Control enabled (optionally named) for control from claude.ai. Short: --rc; requires a claude.ai subscription.

claude --remote-control "laptop session"

claude --remote-control-session-name-prefix <prefix> — Set the prefix for auto-generated Remote Control session names (default: hostname).

claude --remote-control --remote-control-session-name-prefix laptop

claude remote-control — Start a Remote Control server without a local interactive session, to control Claude Code from claude.ai or the Claude app.

claude remote-control

Plugins

claude plugin list — List all installed plugins (claude plugins works as an alias for the whole command group).

claude plugin list

claude plugin install <plugin> — Install a plugin from the available marketplaces; use plugin@marketplace for a specific one. --marketplace <source> (owner/repo, git or https URL, or a path) installs the bare plugin name from that marketplace, adding the marketplace to your user settings first if needed, under the same checks as claude plugin marketplace add (v2.1.292 or later). Alias: i

claude plugin install my-plugin --marketplace your-org/claude-plugins

claude plugin enable <plugin> — Enable a previously disabled plugin.

claude plugin enable my-plugin

/plugin enable cc-plugin-you-should-know@builtin — Turn on You should know, a built-in mod: a side agent watches longer tasks and shows a note above the prompt when it finds something you or Claude might miss (new in v2.1.287; off by default; first-party sessions with telemetry on, where available for your org).

/plugin enable cc-plugin-you-should-know@builtin

claude plugin disable <plugin> — Disable an installed plugin without removing it.

claude plugin disable my-plugin

claude plugin update <plugin> — Update a plugin to the latest version (restart required).

claude plugin update my-plugin

claude plugin uninstall <plugin> — Uninstall a plugin completely. Alias: remove

claude plugin uninstall my-plugin

claude plugin validate <path> — Validate a plugin or marketplace manifest, or the skills, agents and commands in a directory.

claude plugin validate ./my-plugin/

claude plugin test [dir] — Run a mod's tests: every *.test.ts/*.test.tsx under the folder (default: the current one), importing their kit from claude-code/testing. Runs from your shell with no session, sign-in or network, and exits 1 when a test fails. A mod is a plugin whose JavaScript/TypeScript event handlers change how Claude Code looks and behaves (mods are new in v2.1.287).

claude plugin test ./my-mod

claude plugin eval [target] — Run a plugin's eval cases (evals/**/case.yaml or prompt.md + graders/*.md) on your machine, as you, and report scored results, by default against a no-plugin baseline (each case runs 3 times; the HTML report goes to claude.ai unless you pass --no-publish). The target is a path, a plugin name or plugin@marketplace. Only evaluate plugins you trust; --trust-plugin answers the first-run prompt in CI.

claude plugin eval ./my-plugin

claude plugin init <name> — Scaffold a new plugin at ~/.claude/skills/<name>/; it loads in the next session as <name>@skills-dir. Alias: new

claude plugin init my-helpers

claude plugin details <name> — Show a plugin's component inventory and projected token cost.

claude plugin details my-plugin

claude plugin configure <plugin> — Show a plugin's options and which are unset; --values-stdin saves new values from stdin (new in v2.1.285).

claude plugin configure my-plugin

claude plugin marketplace add <source> — Add a marketplace from a URL, path or GitHub repo; list, update and remove manage the configured ones.

claude plugin marketplace add your-org/claude-plugins

claude plugin prune — Remove auto-installed dependencies that are no longer needed. Alias: autoremove

claude plugin prune

claude --plugin-dir <path> — Load a local plugin directory or .zip for this session only; a folder of plugins loads each child (repeatable).

claude --plugin-dir ./my-plugin/ --plugin-dir ./other.zip

claude --plugin-url <url> — Fetch a plugin .zip from a URL for this session only (repeatable).

claude --plugin-url https://example.com/my-plugin.zip

Auto Mode

claude auto-mode config — Print the effective auto mode config as JSON: your settings where set, defaults otherwise.

claude auto-mode config

claude auto-mode defaults — Print the shipped default auto mode rules as JSON (environment, allow, soft_deny, hard_deny).

claude auto-mode defaults

claude auto-mode critique — Get AI feedback on your custom auto mode rules.

claude auto-mode critique

claude auto-mode reset — Reset auto mode to the shipped defaults by removing the autoMode section from your user settings file; asks for confirmation, -y skips it.

claude auto-mode reset

/auto-mode-setup — Draft autoMode.environment entries from your project and recent sessions, review the draft and save it to your user settings (Pro, Max or Team plan).

/auto-mode-setup

Debugging & Diagnostics

claude --debug — Enable debug mode: verbose logs, tool calls, API requests.

claude --debug

claude --debug=<filter> — Filter debug output by category; ! excludes one (--debug='!1p,!file'). The filter only binds in the = form — --debug api,hooks turns on debug mode without filtering.

claude --debug=api,hooks

claude --debug-file <path> — Write debug logs to a file (implicitly enables debug mode).

claude --debug-file /tmp/claude-debug.log

claude --verbose — Enable verbose output mode.

claude --verbose

claude --safe-mode — Start with all customizations disabled (CLAUDE.md, skills, installed plugins, hooks, MCP servers, custom commands and agents, output styles, workflows, custom themes, keybindings) to troubleshoot a broken configuration. Policy settings, auth, model selection, built-in tools and plugins, and permissions still work. Sets CLAUDE_CODE_SAFE_MODE=1.

claude --safe-mode

claude --ax-screen-reader — Render screen-reader friendly output (flat text, no decorative borders or animations).

claude --ax-screen-reader

Scripting & Automation

git diff HEAD~1 | claude -p "Summarize these changes" — Pipe git diff into Claude to generate a commit message.

git diff HEAD~1 | claude -p "Write a conventional commit message"

claude -p --output-format json "<prompt>" | jq '.result' — Pipe JSON output to jq for processing in shell scripts.

claude -p --output-format json "Extract function names" | jq '.result'

claude --settings '<json>' — Load settings from a JSON file path or inline JSON string.

claude --settings '{"model":"sonnet","verbose":true}'

claude --setting-sources <sources> — Control which setting sources are loaded: user, project, local.

claude --setting-sources user,project

claude -p --betas <feature> "<prompt>" — Enable a beta feature via API header (API key users only); use the exact, dated header name from the API docs.

claude -p --betas <beta-header-name> "Think step by step"

claude --disable-slash-commands — Disable all skills for this session.

claude --disable-slash-commands

claude --chrome — Enable Claude in Chrome integration for this session.

claude --chrome

claude --no-chrome — Disable the Claude in Chrome integration for this session.

claude --no-chrome

claude --file <file_id:path> — Download file resources at startup. Format: file_id:relative_path (repeatable).

claude --file file_abc:doc.txt --file file_def:img.png

claude --exclude-dynamic-system-prompt-sections — Move per-machine sections (cwd, env, memory paths, git status) into the first user message for better cross-user prompt-cache reuse.

claude --exclude-dynamic-system-prompt-sections -p "Review"

Environment Variables

CLAUDE_CONFIG_DIR=<path> claude — Use a custom config directory instead of the default ~/.claude. Enables separate configs for different projects or contexts.

CLAUDE_CONFIG_DIR=~/.claude-work claude

alias claude-<name>="CLAUDE_CONFIG_DIR=~/.<dir> claude" — Create shell aliases for different Claude configurations (add to .bashrc/.zshrc).

alias claude-private="CLAUDE_CONFIG_DIR=~/.claude-private claude"
alias claude-work="CLAUDE_CONFIG_DIR=~/.claude-work claude"

ANTHROPIC_API_KEY=<key> claude — Set the API key directly instead of using OAuth login.

ANTHROPIC_API_KEY=sk-ant-xxx claude -p "Hello"

ANTHROPIC_MODEL=<model> claude — Set the model without the --model flag — an alias or a full model name.

ANTHROPIC_MODEL=sonnet claude

ANTHROPIC_DEFAULT_MODEL=<model> — Choose the model new sessions start on by default (v2.1.236 or later) — only when no --model, ANTHROPIC_MODEL or saved model setting (e.g. from /model) applies.

export ANTHROPIC_DEFAULT_MODEL=opus

ANTHROPIC_DEFAULT_OPUS_MODEL=<model-id> — Pin what the opus alias resolves to; ANTHROPIC_DEFAULT_SONNET_MODEL, …_HAIKU_MODEL and …_FABLE_MODEL work the same way. …_HAIKU_MODEL also sets the model for background tasks (replaces the deprecated ANTHROPIC_SMALL_FAST_MODEL).

ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5 claude

CLAUDE_CODE_DISABLE_1M_CONTEXT=1 claude — Turn off 1M-context support: [1m] variants leave the model picker, and models with a native 1M window (e.g. Opus 4.7+, Sonnet 5.5, Fable) are held to a 200K window (compacting at 200K, or stopping with the context-limit error when auto-compaction is off). This also undoes the 1M default that Opus 4.7+ and Fable get on Bedrock, Vertex, Foundry and the Claude apps gateway since v2.1.287.

CLAUDE_CODE_DISABLE_1M_CONTEXT=1 claude --model opus

CLAUDE_CODE_SIMPLE=1 claude — Minimal system prompt and only the Bash, file read and file edit tools; also set by --bare. OAuth login isn't read — authenticate with ANTHROPIC_API_KEY or an apiKeyHelper.

CLAUDE_CODE_SIMPLE=1 claude -p "Quick question"

ANTHROPIC_BASE_URL=<url> claude — Use a custom API endpoint (proxy or compatible provider).

ANTHROPIC_BASE_URL=https://my-proxy.example.com claude

CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS=<ms> — Set a longer base delay (ms) for the backoff when Claude Code retries a request the API rejected as overloaded (529) (v2.1.292 or later).

CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS=5000 claude -p "Summarize this repo"

CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS=1 — Stop sending the structured-outputs request field (and its beta header), for Mantle or an LLM gateway whose upstream rejects it. Without it, session titles, memory recall and prompt hooks fail there (v2.1.288 or later).

CLAUDE_CODE_DISABLE_STRUCTURED_OUTPUTS=1 claude

MCP_PROTOCOL_NEGOTIATION=legacy claude — Keep every MCP server on the earlier handshake instead of probing for protocol revision 2026-07-28, which stdio and HTTP servers are asked for by default on the v2 MCP runtime (stdio on every install since v2.1.292). Use it when an older server stops connecting. auto probes HTTP, claude.ai connector and stdio servers.

MCP_PROTOCOL_NEGOTIATION=legacy claude

CLAUDE_CODE_WEB_SEARCH_REFILLS_PER_HOUR=<n> — Set how fast a session's WebSearch cap (200 calls by default) refills, in calls per hour. The default is 100 in interactive sessions and 0 (no refill) in non-interactive -p runs; 0 turns refill off (v2.1.290 or later; before that the cap never refilled).

CLAUDE_CODE_WEB_SEARCH_REFILLS_PER_HOUR=300 claude

DISABLE_AUTOUPDATER=1 claude — Disable automatic background updates; a manual claude update still works. DISABLE_UPDATES=1 blocks both.

DISABLE_AUTOUPDATER=1 claude

CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR=1 — Return to the original working directory after each Bash or PowerShell command in the main session.

CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR=1 claude

CLAUDE_CODE_NO_FLICKER=1 claude — Enable fullscreen rendering (research preview): less flicker and flat memory in long conversations; same as /tui fullscreen.

CLAUDE_CODE_NO_FLICKER=1 claude

Useful Aliases & Tips

alias claude-private="CLAUDE_CONFIG_DIR=~/.claude-private claude" — Separate config for creative/personal projects — own settings, memory, and MCP servers.

# In ~/.bashrc or ~/.zshrc:
alias claude-private="CLAUDE_CONFIG_DIR=~/.claude-private claude"

alias claude-work="CLAUDE_CONFIG_DIR=~/.claude-work claude" — Separate config for work projects — own API key, permissions, and agents.

# In ~/.bashrc or ~/.zshrc:
alias claude-work="CLAUDE_CONFIG_DIR=~/.claude-work claude"

alias cq='claude -p' — Quick alias for non-interactive one-shot queries.

cq "What's the regex for an email address?"

alias cr='claude -c' — Quick alias to continue the last conversation.

cr

alias claude-review='claude -p --model opus "Review this code for bugs and security issues"' — Predefined alias for code review with a specific model.

cat src/auth.py | claude-review

ccommit() { git diff --cached | claude -p "Write a conventional commit message for these staged changes"; } — Shell function to generate commit messages from staged changes.

git add -p && ccommit

cexplain() { cat "$1" | claude -p "Explain this file concisely"; } — Shell function to quickly explain any file.

cexplain src/config.ts

ultracode — Keyword trigger: include the word 'ultracode' in a prompt to run that task as a dynamic workflow without changing the session's effort level; Option+W / Alt+W dismisses the highlight. /effort ultracode turns it on for every substantive task in the session.

ultracode: refactor the auth module and add full test coverage

ultrathink — Keyword: include ultrathink in a prompt to ask for deeper thinking on that turn without changing the effort level.

ultrathink: why does this race condition only show up under load?

Keyboard Shortcuts (interactive session)

Enter — Submit the current prompt (configurable in /keybindings).

Type your prompt and press Enter to send

Shift+Enter — Insert a newline without submitting — native in iTerm2, WezTerm, Ghostty, Kitty, Warp, Apple Terminal and Windows Terminal; elsewhere use \+Enter or Ctrl+J, or run /terminal-setup (VS Code, Cursor, Alacritty, Zed).

Write multi-line prompts with Shift+Enter

Ctrl+Enter — Send now: deliver your queued messages, with your draft, without waiting (v2.1.275+). Running shell commands and subagents move to the background and keep running (v2.1.281+); if Claude is only writing, the turn is interrupted. Ctrl+X Ctrl+S does the same in terminals that don't report Ctrl+Enter.

Type a correction mid-turn and press Ctrl+Enter

Ctrl+C — Interrupt a running operation; with nothing running, the first press clears the input and a second press exits. Pressing Up on the empty prompt brings a cleared draft back, pasted text and images included (v2.1.288 or later; since v2.1.292 also after a slash command or a sent message).

Press Ctrl+C, then Up to restore the draft

Escape — Interrupt Claude mid-turn (the work so far is kept) or close a dialog.

Press Escape to stop and redirect Claude

Escape Escape — Clear the input draft (saved to history), or with an empty input open the rewind menu.

Press Escape twice to rewind to an earlier point

Shift+Tab — Cycle permission modes: Manual, acceptEdits, plan and, where available, bypassPermissions and auto.

Press Shift+Tab until the mode indicator shows plan

Ctrl+R — Reverse-search your prompt history.

Press Ctrl+R and type part of an earlier prompt

Ctrl+O — Toggle the transcript viewer with detailed tool usage.

Press Ctrl+O to see what a tool call did

Ctrl+T — Show or hide Claude's to-do checklist in the status area (for running shells and subagents use /tasks).

Press Ctrl+T to see Claude's task list

Ctrl+B — Send running Bash commands and agents to the background (press twice in tmux).

Press Ctrl+B while a long test run is going

Ctrl+X Ctrl+K — Stop all running background subagents in this session (press twice within 3 seconds to confirm). Works while a background subagent's permission prompt is open; since v2.1.290 those prompts show this shortcut.

Press Ctrl+X Ctrl+K twice to stop every background agent

Ctrl+G — Open the prompt in your default text editor.

Press Ctrl+G to write a long prompt in your editor

Ctrl+S — Stash the prompt: with text in the input it is saved and cleared; press again on an empty prompt to restore it, including cursor position, pasted content and shell mode.

Press Ctrl+S to park a half-written prompt and ask something else first

Ctrl+V — Paste an image from the clipboard as an [Image #N] chip at the cursor (Cmd+V in iTerm2, Alt+V on Windows; on WSL use Alt+V if the terminal intercepts Ctrl+V).

Press Ctrl+V to paste a screenshot into the prompt

Alt+P / Option+P — Switch the model without clearing your prompt.

Press Alt+P (Option+P on macOS) to pick another model

Alt+O / Option+O — Toggle fast mode.

Press Alt+O (Option+O on macOS) to toggle fast mode

Up / Down Arrow — Navigate through prompt history.

Press Up to recall previous prompts

/ — Open slash command autocomplete. Type to filter.

Type / to see all available commands

! — At the start of the prompt: shell mode — run a command directly and add its output to the session.

!git status

@ — Mention a file path with autocomplete.

Explain @src/utils/date.ts

Conclusion

Claude Code rewards a deliberate setup: pin sensitive operations behind permission modes (interactive sessions now start in auto mode where it is available; Shift+Tab or --permission-mode manual brings the prompts back), reserve --dangerously-skip-permissions for throwaway sandboxes only, and lean on -p plus --output-format json when you wire it into scripts and CI. Once you have a feel for sessions, agents and MCP servers, the same binary scales from a quick one-off question to a long-running, multi-step refactor. Treat the flags here as a starting point and check the official docs for new commands between releases.

Further Reading

• Claude Code documentation (opens in a new tab) – Anthropic's official guide to installation, configuration, slash commands, hooks and MCP
• Commands reference (opens in a new tab) – the full list of built-in slash commands and bundled skills
• CLI reference (opens in a new tab) – every command-line command and option
• Model configuration (opens in a new tab) – aliases, model versions per provider and environment variables for model selection
• anthropics/claude-code on GitHub (opens in a new tab) – issue tracker, release notes and reference material