Commands
Every built-in command, bundled skill and bundled workflow you can run with / in Claude Code, grouped by what you are trying to do.
Commands are how you steer Claude Code from inside a session: switch model, tidy context, change permissions, review a diff, hand work to the background. Type / to see what is available to you and keep typing to filter.
A few ground rules:
- A command only counts at the start of a message. Whatever follows the name is its argument.
- Skills are the exception to "one per message": you can chain up to six at the start, as in
/tdd /house-style add a VAT calculator, and the trailing text goes to each of them. - If you send a command while Claude is mid-response, it is queued until the turn ends. Some run immediately without interrupting, such as
/status,/tasksand/usage, and in fullscreen rendering dialog commands like/themeand/helpopen straight away too (from v2.1.234).
Availability varies with platform, plan and provider. /desktop, for example, only appears on macOS and x64 Windows with a Claude subscription, and /upgrade is hidden on Enterprise. Your own commands are written as skills.
A typical day, in commands
To show how these fit together, here is roughly how I use them on a client project:
- New repo.
/initfor a starterCLAUDE.md,/memoryto tighten it,/mcpto connect the servers the project needs,/permissionsto set allow rules. - Mid-task.
/planbefore anything big./modeland/effortwhen I want more (or less) horsepower./contextwhen things feel sluggish, then/compact./btwfor a quick question I do not want in the history. - Parallel work.
/tasksto see background work,/backgroundto detach the whole session,/batchfor a sweeping change split across worktrees. - Before shipping.
/diff, then/code-review(with--fixif I trust it), and/security-reviewon anything touching auth. - Between tasks.
/clearfor a fresh start,/resumeto go back,/branchto try an alternative. - When it goes wrong.
/rewind,/doctor,/debug, and/bugif it is genuinely broken.
Reference
<arg> is required, [arg] optional. Rows marked (skill) are bundled skills: prompts handed to Claude rather than fixed logic. Rows marked (workflow) are bundled dynamic workflows that fan out across subagents in the background. /verify and /deep-research only run when you invoke them.
Sessions and conversation
| Command | What it does |
|---|---|
/clear [name] | Start a fresh conversation with empty context. A name labels the old one in the /resume picker. The previous conversation can be resumed, or recovered from the rewind menu in the same process. Aliases /reset, /new |
/resume [session] | Resume by id or name, or open the picker. Background sessions show as bg; resuming a running one attaches this terminal to it and moves your current conversation to the background (press ← on an empty prompt to return to agent view). Alias /continue |
/branch [name] | Branch the conversation at this point and switch into the branch; the original stays available via /resume |
/fork [prompt] | Copy the conversation into a new background session while you carry on here. With a prompt the copy starts work immediately. Unless it edits in place, the copy is told to make its own worktree first. Needs v2.1.212; when agent view is off it starts a forked subagent instead |
/subtask <task> | Start a forked subagent that inherits the whole conversation, works in the background, and reports back here. v2.1.212 or later; unavailable when agent view is off |
/rename [name] | Rename the session (shown on the prompt bar). No name: one is generated from the conversation. Control and invisible characters become spaces and names are capped at 200 characters. Clashing names get a variant. Works in -p |
/rewind | Roll code and/or conversation back to a checkpoint, or summarise from a message. See Checkpointing. Aliases /checkpoint, /undo |
/export [filename] | Export the conversation as plain text, to a file or via a copy/save dialog |
/copy [N] | Copy the last (or Nth-latest) response. With code blocks, a picker lets you choose blocks; press w to write to a file instead, handy over SSH |
/recap | One-line summary of the session on demand |
/cd <path> | Move the session to another working directory, keeping the conversation. Tab completes paths. See Permissions for how it differs from /add-dir |
/add-dir <path> | Grant file access to another directory for this session. Most .claude/ config there is not discovered, and most network paths are refused. Runs DirectoryAdded hooks. Mid-turn, you confirm immediately and Claude can use it on its next tool call |
/exit | Quit. In an attached background session it detaches instead. Alias /quit |
Context, memory and model
| Command | What it does |
|---|---|
/context [all] | Coloured grid of context usage with suggestions and capacity warnings. In fullscreen, per-item detail is collapsed unless you pass all |
/compact [instructions] | Summarise the conversation to free space, optionally with a focus. See Context window |
/autocompact [auto|<tokens>] | Set how full context gets before auto-compaction, e.g. 500k, or auto for the model-tuned default. Saved to user settings. No argument opens a dialog. v2.1.221 or later |
/memory | Edit CLAUDE.md files, toggle auto memory, browse memory entries |
/init | Generate a starter CLAUDE.md. CLAUDE_CODE_NEW_INIT=1 gives an interactive flow covering skills, hooks and personal memory. Offers /import if it finds Codex or Gemini CLI config |
/model [model] | Switch model and save it as your default. Left/right arrows adjust effort where supported; s in the picker switches for this session only. Applies mid-turn after any confirmation. In -p, takes a model argument and applies to the session only. See Model configuration |
/effort [level|auto|status|ultracode [on|off]] | Set effort low to xhigh, max (session only) or auto; status prints it. ultracode toggles ultracode at the current level (v2.1.284). Applies to the next request in the turn after any cache warning. Works in -p |
/fast [on|off] | Toggle fast mode; the running turn finishes at its original speed. v2.1.205 or later |
/advisor [model|off] | Turn the advisor on or off: fable, opus, sonnet or a model id. No argument opens a picker (or prints the current advisor where there is no terminal) |
/output-style [style] | List or switch output styles. v2.1.269 or later |
/btw [question] | Ask a side question without adding to the conversation. No argument shows recent side answers |
/goal [condition|clear] | Keep Claude working across turns until a condition is met (Goal). No argument shows the current goal; clear, stop, off, reset, none or cancel removes it |
/plan [description] | Enter plan mode, optionally starting on a task: /plan tidy up the checkout flow |
Permissions, safety and review
| Command | What it does |
|---|---|
/permissions | Manage allow, ask and deny rules by scope, working directories, recent auto mode denials, and auto mode classifier rules on the Auto mode tab. Opens immediately mid-turn. Alias /allowed-tools |
/sandbox | Toggle sandboxing on supported platforms |
/auto-mode-setup | Draft autoMode.environment entries from your project and recent sessions, review and save to user settings. Pro, Max or Team; v2.1.228 or later. See Auto mode configuration |
/fewer-permission-prompts | (skill) Mine your transcripts for common read-only Bash and MCP calls and add a prioritised allowlist to project .claude/settings.json |
/diff | Review working-tree changes, including Claude's edits so far |
/code-review [low|medium|high|xhigh|max|ultra] [--fix] [--comment] [--max-findings n|all|default] [pr#|branch|path] | (skill) Review the current diff, or a PR, branch or path, for correctness bugs (plus clean-ups at some effort levels). --fix applies findings, --comment posts them to the GitHub PR or GitLab MR, ultra runs a cloud ultrareview, and with ultra on a github.com PR --post preselects posting. Alias /review. See Code review |
/review ... | Alias of /code-review with the same arguments; with no level it reuses the last level you typed |
/ultrareview [PR or branch] | Deep multi-agent review in a cloud sandbox; prefer /code-review ultra. Three free runs on Pro and Max, then usage credits |
/security-review | Check your branch's diff against origin's default branch for vulnerabilities. Needs an origin remote |
/simplify [target] | (skill) Four parallel agents look for reuse, simplification, efficiency and abstraction level, then apply fixes. Does not hunt bugs |
Parallel and background work
| Command | What it does |
|---|---|
/tasks | View and manage background work in this session, including finished subagents. Also /bashes |
/background [prompt] | Detach this session to run as a background agent and free the terminal; monitor with claude agents. Alias /bg |
/stop | Stop the background session you are attached to (or a peek-reply target); transcript and worktree are kept |
/batch <instruction> | (skill) Plan a codebase-wide change as 5 to 30 independent units, then run one background subagent per unit in its own worktree, each testing and publishing its change. Needs git or a WorktreeCreate hook (outside git, v2.1.281 or later) |
/loop [interval] [prompt] | (skill) Repeat a prompt while the session is open, e.g. /loop 10m check the staging deploy. No interval: Claude self-paces. No prompt: the built-in maintenance prompt or your loop.md. Alias /proactive. See Scheduled tasks |
/schedule [description] | Create, update, list or run cloud routines conversationally. Alias /routines |
/list-agents | List subagents, teammates and other sessions Claude can message, with their names. Also /peers. Only where cross-session messaging is on; v2.1.224 or later |
/workflows | Watch, pause, resume or save running and finished workflows |
/deep-research <question> | (workflow) Fan out web searches, cross-check sources and produce a cited report |
Configuration and interface
| Command | What it does |
|---|---|
/config [key=value ...] | Open settings, or set keys directly (/config theme=dark, /config model=sonnet). The direct form works in -p and over Remote Control but cannot switch on settings that need in-panel confirmation. /config --help lists keys. Alias /settings |
/status | Settings on the Status tab: version, model, account, connectivity, and a Session kind row. Works mid-response |
/theme | Pick a colour theme, including auto, daltonised and ANSI themes, and custom themes from ~/.claude/themes/ or plugins |
/color [color|default] | Prompt bar colour for this session: red, blue, green, yellow, purple, orange, pink, cyan. No argument picks one at random. Syncs to claude.ai/code under Remote Control |
/tui [default|fullscreen] | Choose the renderer and relaunch with the conversation intact |
/focus | Toggle the focus view (last prompt, one-line tool summary, final response). Fullscreen only; persists, overridable with viewMode |
/scroll-speed | Tune mouse-wheel scrolling in fullscreen (not in JetBrains' terminal) |
/statusline | Configure the status line, by description or from your shell prompt |
/keybindings | Open your keybindings file |
/terminal-setup | Install Shift+Enter for newlines in VS Code, Cursor, Devin Desktop, Alacritty or Zed; in Apple Terminal enable Option+Enter and silence the bell; in iTerm2 allow clipboard access for /copy. See Terminal configuration |
/voice [hold|tap|off] | Toggle voice dictation. Needs a Claude.ai account |
/update-config [request] | (skill) Describe a settings change (allow a command, set an env var, add a hook) and Claude edits the right settings.json |
/import [codex|gemini|cursor] [--dry-run] [--yes] | Bring over instructions, MCP servers, commands, subagents and skills from Codex, Gemini CLI or Cursor. Not on Bedrock, Agent Platform, Foundry, Claude Platform on AWS, gateways, or without feature-flag fetching. v2.1.213 (Cursor v2.1.265) |
/remote-env | Choose the default cloud environment for cloud sessions started from the CLI |
Plugins, skills, MCP and hooks
| Command | What it does |
|---|---|
/plugin [subcommand] | Open the plugin panel or act directly (list, install, enable, disable...). See the plugin CLI reference |
/reload-plugins [--force] | Apply pending plugin changes without restarting; refuses changes that would invalidate the prompt cache unless forced |
/plugin-authoring | Load the reference Claude uses to write a mod. From a built-in plugin; v2.1.287 or later |
/skills | List skills; filter by typing, t sorts by tokens, Space/Enter cycles visibility (not for plugin skills, disable-model-invocation skills, or managed overrides) |
/reload-skills | Rescan skill and command folders for changes made during the session |
/skill-doctor | Show each skill's context cost and usage to find ones to turn off. v2.1.252 and feature-flag fetching |
/mcp [reconnect (<server>|all)|enable|disable [<server>|all]] | Manage MCP connections and OAuth. reconnect all retries failed or unauthenticated servers. In -p, no argument prints a status summary |
/hooks | View hook configuration |
/agents | Reminds you to ask Claude to create subagents or edit .claude/agents/ directly (older versions had an interactive editor) |
MCP servers can also expose prompts, which appear as extra commands; see MCP.
Integrations and platforms
| Command | What it does |
|---|---|
/ide | Manage IDE integrations and show status |
/chrome | Configure Claude in Chrome |
/claude-in-chrome [task] | (skill) Carry out a browser task (test a page, fill a form, read the console) via Claude in Chrome |
/desktop | Continue this session in the desktop app. macOS or x64 Windows, subscription required. Alias /app |
/remote-control | Make this session reachable via Remote Control. Alias /rc |
/teleport | Pull a cloud session into this terminal. Also /tp. Subscription required |
/web-setup | Connect GitHub for cloud sessions using your local gh credentials |
/autofix-pr [prompt] | Start a cloud session that watches the current branch's PR and pushes fixes for CI failures and review comments, or follows your narrower instruction. Needs gh and cloud access |
/install-github-app | Install the Claude GitHub App and optionally set up GitHub Actions. github.com only |
/install-slack-app | Install the Claude Slack app via browser OAuth |
/mobile | QR code for the mobile app. Aliases /ios, /android |
/setup-bedrock | Wizard for Amazon Bedrock auth, region and model pins. Hidden until CLAUDE_CODE_USE_BEDROCK=1 |
/setup-vertex | Wizard for Google Cloud's Agent Platform. Hidden until CLAUDE_CODE_USE_VERTEX=1 |
Artifacts and design
| Command | What it does |
|---|---|
/artifacts | List artifacts you own or that are shared with you; attach, open or copy a link. v2.1.208 or later |
/artifact-capabilities | (skill) Reference for runtime capabilities a published artifact can use |
/artifact-diagramming | (skill) Diagramming guidance for artifacts, with inline SVG that works in both themes |
/dataviz [request] | (skill) Chart and dashboard design guidance with a colour-safe placeholder palette |
/design [brief] | (skill) Draft mockups or posters as artboards on a Claude Design artifact. Needs v2.1.265, artifacts, and the Design template; not on third-party providers |
/design-login | Authorise design-system access for /design-sync |
/design-sync [hint] | (skill) Convert your React design system and upload it to Claude Design. First syncs can take hours |
/slides [brief] | (skill) Build a Claude Slides deck from a brief. Same requirements as /design with the Slides template |
Running and verifying
| Command | What it does |
|---|---|
/run | (skill) Launch and drive your app to see a change actually working |
/verify | (skill) Build, run and observe the app to confirm a change behaves correctly |
/run-skill-generator | (skill) Write a per-project skill teaching /run and /verify how to build and launch your app |
/claude-api [migrate|upgrade|managed-agents-onboard|prompt-audit|cost-optimize|build-eval|hillclimb|preserved-thinking-migration] | (skill) Claude API and Managed Agents reference for your language; activates automatically when your code imports the Anthropic SDK |
/workflow-authoring | (skill) Reference for writing workflow scripts. v2.1.248 or later |
Account, usage and help
| Command | What it does |
|---|---|
/login, /logout | Sign in or out of your Anthropic account |
/usage | Session cost, plan limits and activity stats, with a breakdown of what counts against your plan. Aliases /cost, /stats (opens on Stats) |
/usage-credits | Configure usage credits, or (Team/Enterprise without billing access) request them from your admin. Formerly /extra-usage |
/rate-limit-options | Ways to keep going when a usage limit bites: wait and auto-continue, add credits, or upgrade |
/upgrade | Open the upgrade page |
/privacy-settings | View and change privacy settings (Pro and Max) |
/insights | HTML report on your recent sessions on this machine. Not in cloud sessions. See Costs |
/team-onboarding | Turn your last 30 days of usage into an onboarding guide for a teammate, with a share link for subscribers |
/passes | Share a free week of Claude Code, if eligible |
/help | Help and command list |
/powerup | Short interactive lessons on features |
/release-notes | Changelog picker; notes appear in the transcript but not in Claude's context |
/doctor [prompt-audit [path]] | (skill) Diagnose installation, settings, extension and CLAUDE.md problems and apply fixes after you confirm; prompt-audit audits your instructions. Alias /checkup. See Debug your configuration |
/debug [description] | (skill) Turn on debug logging from this point and analyse the log |
/heapdump | Write a heap snapshot and memory breakdown to ~/Desktop (or home). Share only the -diagnostics.json; the snapshot contains your conversation and credentials. Hidden from the menu |
/bug [report] | Report a bug, choosing how much history to include after a consent screen. Goes to Anthropic on first-party connections, otherwise to a local archive under ~/.claude/feedback-bundles/. Alias /share |
/feedback [report] | Product feedback via the same dialog; with Claude-drafted feedback enabled, no argument opens the drafts queue |
/stickers, /radio | Order stickers; open the lo-fi radio stream |
Removed commands
/pr-comments (removed in v2.1.91; just ask Claude), /vim (removed in v2.1.92; use /config → Editor mode) and /ultraplan (removed; use plan mode).
How the / menu matches
- Highlighting. The top suggestion is highlighted only when your letters match a command name or alias from its start or from a word inside it, ignoring
:,_and-./adddirhighlights/add-dir;/newhighlights/clearvia its alias.Enterruns it (v2.1.236 or later). - Typos. Nothing is highlighted. Close matches stay listed for
Tabor arrow selection, butEntersends what you typed and you getUnknown command(see Errors). - Unavailable commands are left out; with no matches you see
No commands match "/name". Submitting one usually givesUnknown command, though some reply with their own availability or policy message. - Hidden commands such as
/heapdumpnever appear for partial input. They show only once the full name is typed, and running the full name works.