Claude Code
Reference

Starts a new conversation with empty context; the previous one stays available in /resume. Pass a name to label it.

aliases: /reset /new

Compresses the conversation into a summary to reclaim context space. Optional instructions guide what gets preserved.

Picks up a previous conversation by session ID or name, or opens a picker. Background sessions appear marked bg.

alias: /continue

Branches the conversation here so you can try a different direction; switches you in and keeps the original, returnable via /resume.

Spawns a forked subagent that inherits the full conversation and works on the directive in the background while you keep going; its result returns when done. Redefined in v2.1.161 (was an alias for /branch).

Gives the session a custom name shown on the prompt bar, or auto-generates one from conversation history.

Rolls conversation and/or code back to a previous checkpoint, or summarizes from a selected message.

aliases: /checkpoint /undo

Closes the session and returns you to your shell. In an attached background session, detaches and keeps it running.

alias: /quit

Displays a colored grid of context-window consumption with optimization hints. Pass all to expand the per-item breakdown.

Shows session cost plus plan limits and activity stats; on paid plans, breaks usage down by skill, subagent, plugin, and MCP server.

aliases: /cost /stats

Opens usage focused on session cost — input/output tokens and cache hits. Alias for /usage.

Opens usage on the Stats tab — daily patterns, session history, and model preferences. Alias for /usage.

Configures usage credits so Claude keeps working after you hit a limit. Renamed from /extra-usage.

Asks a quick question with full visibility into the session, answered in a dismissible overlay without adding to history. Works while Claude is busy.

Generates a one-line summary of the session so far on demand.

Switches the active model and saves it as the default for new sessions; use arrow keys to adjust effort, or press s for this session only.

Sets reasoning effort: low, medium, high, xhigh, max, or ultracode (max and ultracode are session-only). auto resets to the model default.

Toggles fast mode: same model, optimized for faster output at the cost of some thoroughness.

Enables or disables the advisor, which consults a second model (opus, sonnet, or fable) for guidance at key moments. (v2.1.98+)

Toggles voice dictation for prompts; hold-to-record by default, or /voice tap for tap-to-toggle. Requires a Claude.ai account.

Opens the Settings Status tab — version, model, account, and connectivity. Works even while Claude is responding.

Opens the interactive settings UI, or set a value directly (e.g. /config thinking=false). Run /config help to list keys.

alias: /settings

Manages allow / ask / deny rules for tools by scope, plus working directories and recent auto-mode denials.

alias: /allowed-tools

Manages shell commands that run automatically in response to tool events (pre/post hooks).

Opens CLAUDE.md memory files for editing and manages auto-memory settings.

Creates a CLAUDE.md file in the project root with starter instructions tailored to your codebase.

Opens keybindings config to customize keyboard shortcuts and add chord bindings.

Configures the persistent status bar showing model, context, and live info.

Changes color theme between light, dark, colorblind-accessible, and ANSI palettes.

Configures terminal keybindings for Shift+Enter multi-line input and other shortcuts.

Toggles sandboxed execution mode that restricts file system and network access.

View and update data privacy preferences (Pro and Max subscribers).

[Skill] Scans your transcripts for common read-only Bash/MCP calls and adds an allowlist to project .claude/settings.json to cut permission prompts.

Manages MCP server connections and OAuth; reconnect a server, or enable/disable one or all without opening the dialog.

Installs, removes, and manages plugins. Pass list, install, enable, or disable to act directly.

Hot-reloads active plugins — skills, agents, hooks, plugin MCP/LSP servers — without restarting.

Re-scans skill and command directories so files added or changed on disk become available mid-session. (v2.1.152+)

Lists available skills; press t to sort by token count, Space to hide a skill. Plugin skills appear here too.

Manages subagent configurations — specialized agents with their own tools, model, and instructions.

Manages IDE integrations and shows connection status (VS Code, Cursor, JetBrains).

Configures the Claude in Chrome integration for browser-based workflows and debugging.

Creates, lists, or runs routines that execute on Anthropic-managed cloud infrastructure, set up conversationally.

alias: /routines

[Skill] Runs a prompt repeatedly while the session stays open; omit the interval and Claude self-paces between iterations.

alias: /proactive

[Skill] Loads Claude API / Agent SDK reference for your language; migrate upgrades existing API code to a newer model.

[Workflow] Fans out web searches, fetches and cross-checks sources, and synthesizes a cited report.

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

Interactive diff viewer for uncommitted changes and per-turn diffs; arrows switch between the git diff and individual turns.

Enters plan mode — Claude explores and designs an approach for approval before editing. Pass a task to start immediately.

[Skill] Reviews the diff for correctness bugs and cleanups; --fix applies findings, --comment posts inline PR comments, ultra runs a deep cloud review.

Reviews a pull request locally in the current session.

Scans pending changes on the branch for security vulnerabilities — injection, auth issues, and data exposure.

[Skill] Four parallel agents review for reuse, simplification, efficiency, and abstraction, and apply cleanup fixes (no bug-hunting). (v2.1.154+)

Runs a deep, multi-agent code review in a cloud sandbox. Preferred form is /code-review ultra; /ultrareview is an alias.

Drafts a plan in an ultraplan session you review in the browser, then executes remotely or sends it back to your terminal.

[Skill] Decomposes a codebase-wide change into 5–30 units, runs one background subagent per unit in its own git worktree, and opens a PR each. Requires git.

[Skill] Launches and drives your project's app to see a change working in the running app, not just tests. (v2.1.145+)

[Skill] Confirms a change works by building and running the app and observing the result, rather than relying on tests. (v2.1.145+)

Spawns a web session that watches the current branch's PR and pushes fixes when CI fails or reviewers comment. Requires gh.

Copies Claude's last response to clipboard; pass N for an earlier one, or pick individual code blocks (press w to write to a file).

Exports the full conversation as plain text to clipboard or a specified file.

Makes this session controllable from claude.ai so you can continue it from another device.

alias: /rc

Chooses the default cloud-agent environment for teleporting sessions to cloud dev machines.

Pulls a Claude Code on the web session into this terminal — picks a session, fetches its branch and conversation. Subscription required.

alias: /tp

Connects your GitHub account to Claude Code on the web using your local gh credentials.

Continues the current session in the Claude Code desktop app (macOS/Windows, subscription).

alias: /app

Signs in to your Anthropic account to authenticate Claude Code.

Signs out and clears stored credentials from this machine.

Opens the upgrade page to switch to a higher plan tier. Shown on Pro and Max plans.

Shares a free week of Claude Code with friends through a referral link. Shown if your account is eligible.

Overview of available commands, usage tips, and documentation links.

Diagnoses your install and settings; press f to have Claude fix any reported issues.

[Skill] Enables debug logging mid-session and troubleshoots by reading the session debug log. Off by default unless you started with --debug.

Detaches the current session to run as a background agent and frees this terminal. Pass a prompt to send one more instruction first.

alias: /bg

Stops the current background session; the transcript and any worktree are kept. Only available while attached.

Views and manages everything running in the background — agents and async shell commands.

alias: /bashes

Adds an additional working directory so Claude can access files outside the launch directory.

Moves this session to a new working directory, preserving the prompt cache by appending the new CLAUDE.md. (v2.1.169+)

Toggles focus view — only your last prompt, a one-line tool summary with diffstats, and the final response. Fullscreen only.

Sets the terminal UI renderer and relaunches into it with your conversation intact. fullscreen enables the alt-screen renderer.

Sets the prompt-bar color for the session; run with no argument for a random color.

Generates a report on your sessions — project areas, interaction patterns, and friction points.

Generates a team onboarding guide from your last 30 days of usage that a teammate can paste in to get set up.

Discovers Claude Code features through quick interactive lessons with animated demos.

Opens Claude FM lo-fi radio in your browser; prints the stream URL when no browser is available.

Submits feedback, reports a bug, or shares your conversation with session context.

aliases: /bug /share

Views the changelog in an interactive version picker.

Sets up the Claude GitHub Actions app for CI/CD integration and automated code review.

Installs the Claude Slack app via OAuth to interact with Claude from Slack channels.

Order physical Claude Code stickers for your laptop or workspace.

Displays a QR code to download the Claude mobile app.

aliases: /ios /android

Use /plugin to browse marketplaces and install, enable, or disable plugins. For local development, load a directory with --plugin-dir (repeatable) or a hosted .zip with --plugin-url.

The manifest lives at .claude-plugin/plugin.json (name, description, version). Everything else — skills/, commands/, agents/, hooks/, .mcp.json, .lsp.json — sits at the plugin root, not inside .claude-plugin/.

After editing plugin files, run /reload-plugins to reload skills, agents, hooks, and plugin MCP/LSP servers without restarting. Pass --force if the reload would invalidate the prompt cache.

Anthropic runs two public marketplaces: claude-plugins-official (curated, auto-registered) and the community catalog, added with /plugin marketplace add anthropics/claude-plugins-community and installed from as @claude-community.

Plugins bundle MCP servers via a root .mcp.json (or inline in plugin.json). Plugin skills are namespaced as /plugin-name:skill so they never collide with your own.

Run claude plugin init <name> to scaffold a manifest and starter skill, then claude plugin validate before sharing. Or convert an existing .claude/ setup into a plugin directory.

Personal skills go in ~/.claude/skills/<name>/SKILL.md (all projects); project skills in .claude/skills/<name>/SKILL.md (this repo). Enterprise > personal > project on name clashes.

Only description is recommended — Claude uses it to decide when to load the skill. Optional fields include name, allowed-tools, model, effort, argument-hint, and paths. The Markdown body is the prompt.

By default both you and Claude can invoke any skill. Set disable-model-invocation: true so only you can run it (good for /deploy), or user-invocable: false to hide it from the / menu so only Claude loads it.

allowed-tools grants tools without a permission prompt while the skill is active; disallowed-tools removes tools from Claude's pool — useful for autonomous loops that should never call AskUserQuestion.

A !`command` line runs a shell command and inlines its output before Claude sees the skill, so !`git diff HEAD` arrives with the live diff already filled in. $ARGUMENTS (or $0, $1) captures input.

Use /skills to see everything, or type / and filter. Edits to SKILL.md apply live; new top-level skill directories need a restart, or /reload-skills to re-scan.

HTTP is preferred for hosted services: claude mcp add --transport http notion https://mcp.notion.com/mcp. Add --header "Authorization: Bearer …" for token auth. SSE and WebSocket transports also exist.

Run a local process, separating server args with --: claude mcp add --env KEY=val airtable -- npx -y airtable-mcp-server. Everything after -- is passed to the server untouched.

--scope local (default) is just you in this project; --scope project is shared via a .mcp.json file in the repo; --scope user is you across all projects.

For OAuth servers, add the URL then run /mcp in-session to authenticate in the browser. Project-scoped servers from .mcp.json show as ⏸ Pending approval until you approve them once.

Use claude mcp list to see servers, claude mcp get <name> for details, claude mcp remove <name> to delete, or /mcp in-session for status and tool counts.

With tool search on (the default), MCP tool schemas load on demand via the ToolSearch tool, so servers with many tools don't bloat the context window. Troubleshoot connections in /mcp; pass env vars with --env KEY=value.