Cursor Commands & Prompt Reference
Commands typed into Cursor's agent, checked on 2026-10-10 against the CLI slash-command reference, the CLI changelog through release v2026.09.28, and the Prompting, Agent and Skills docs. The first six groups are CLI slash commands, a few of them documented only in the changelog. The last two cover the chat input, where @ mentions attach context and / runs built-in or your own skills; in the CLI, @ picks files and folders. Launch flags and editor keyboard shortcuts are left out.
Also see: Cursor keyboard shortcuts.
All Commands
Models and modes (8)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /model [filter] | Pick a model | Text after the command pre-filters the picker, as in /model sonnet; Auto stays pinned at the top of the list. |
| /fast | Toggle the Fast variant | Works only when the current model offers Fast, and that variant can require Max Mode. |
| /run-everything [on|off|status] | Toggle Run Everything | Alias /auto-run; approval is set here, apart from the Shift+Tab cycle between Agent, Plan and Ask. |
| /plan [prompt] | Plan before coding | Shift+Tab also rotates into Plan mode, where the agent asks clarifying questions before any code is written. |
| /ask | Toggle Ask mode | The agent searches the codebase and answers without editing files; start in it with --mode=ask. |
| /debug [prompt] | Toggle Debug mode | With no prompt it only switches the mode; with one, it sends that prompt in Debug mode. |
| /goal [objective] | Set a long-lived goal | Still rolling out; it also works in the editor chat, and in the CLI Ctrl+C pauses the goal. |
| /max-mode | Toggle Max Mode | Only for legacy request-based plans; the choice persists across sessions and shows MAX in the prompt bar. |
Chat sessions (10)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /clear | Start a new chat | Also accepted as /new, /new-chat or /newchat. |
| /resume | Resume a recent chat | Lists chats from all workspaces by default; Left and Right switch to this workspace, and Ctrl+Y opens it too. |
| /rename <name> | Rename this chat | |
| /fork | Fork the chat | Also /branch or /duplicate; explore another direction while the original conversation stays as it was. |
| /rewind | Go back to an earlier turn | A timeline restores files and conversation to the chosen turn; on by default since v2026.06.22. |
| /copy | Copy an earlier message | A picker can also copy the agent's replies or a single step of a multi-step answer. |
| /btw | Ask a side question | The read-only answer appears in a dismissible overlay mid-turn and never enters the chat history. |
| /detach | Leave a persistent session | Used in a session started with agent persist; the agent keeps working until you reattach with agent persist attach. |
| /quit | Quit the CLI | Same as /exit; pressing Ctrl+D twice on the prompt also leaves the CLI. |
| /exit | Exit the CLI | Returns to the shell at once, without waiting for MCP servers or background tasks to stop. |
Context and workspaces (6)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /summarize | Summarize the conversation | Also /compress or /compact; afterwards your latest message is still there word for word. |
| /context | Show context window use | Breaks the window down by category, so you can see what is taking up space before summarizing. |
| /add-dir | Add a directory | Adds another directory mid-session; its slash skills and custom commands become available at once. |
| /save-workspace | Save a named workspace | Stores the current set of directories so a later session can load the same repositories. |
| /load-workspace | Load a saved workspace | Reloads a directory set saved earlier; the --workspace flag starts a session scoped to one from the shell. |
| /open | Open the repo in Cursor | Alias /cursor; the editor window opens on the top-level folder of the current Git repository. |
Tools and settings (6)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /shell [command] | Enter Shell Mode | Short aliases /sh and /run open the same Shell Mode. |
| /mcp [list|list-tools] [identifier] | Manage MCP servers | Opens a per-server view to log in or out, enable or disable a server and browse its tools; type to filter. |
| /plugin [subcommand] | Manage plugins | Installs or removes plugins at user or project scope; /plugin marketplace add registers a marketplace by git URL. |
| /config | Open the settings editor | Covers permissions such as auto-accepting web searches, model slash shortcuts and the Explore subagent's model. |
| /sandbox | Configure the sandbox | Sandboxed commands reach the network only through a default-deny proxy, plus any allowlists your team admin sets. |
| /bedrock [subcommand] | Set up AWS Bedrock | Only offered once the Bedrock feature is enabled. |
Display and terminal (5)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /vim | Toggle Vim keys | Starts Vim editing from any point in the prompt; Esc returns to normal mode while you revise a plan. |
| /line-numbers | Toggle code line numbers | |
| /show-thinking | Toggle thinking blocks | |
| /status-indicators | Toggle title status indicators | |
| /setup-terminal | Set up newline keys | Shift+Enter inserts a newline in iTerm2, Ghostty, Kitty, Warp and Zed; inside tmux use Ctrl+J instead. |
Help and account (9)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /help [command] | Show help | Name a command after it to read that command's details instead of the general help. |
| /about | Show version and account info | The same details are also copied to the clipboard. |
| /usage | Show plan usage and spend | Lists included-usage meters, on-demand spending versus your cap, the plan name and when the billing cycle resets. |
| /feedback <message> | Send feedback to Cursor | |
| /logs | Find the debug log | Every session writes a debug log, stored per user on shared machines; the path is also copied. |
| /update | Update the CLI | Upgrades in place from inside the session; agent update does the same from the shell. |
| /copy-request-id | Copy last request ID | |
| /copy-conversation-id | Copy the conversation ID | |
| /logout | Sign out |
Mentions and chat input (10)
| Command / Syntax | What it does | Notes |
|---|---|---|
| @auth.ts | Attach a file | auth.ts is the docs' example: type any file name after @, in the editor chat or the CLI, and pick a suggestion. |
| @src/components/ | Attach a folder | After choosing a folder, type / to move into its subfolders before attaching. |
| @Terminals | Attach terminal output | |
| @Chats | Reference a past chat | |
| @Commit | Attach uncommitted changes | Listed in the menu as Diff of Working State; it covers only what you have not committed yet. |
| @Branch | Attach the branch diff | Listed as Diff with Main: every change on your branch compared with main, the usual context for a review. |
| @Browser | Attach browser context | |
| /skill-name | Invoke a skill | Enter attaches it to one message; Alt+Enter (Option+Enter on Mac) or Use as Mode keeps it on every turn. |
| /side | Open a side chat | Also /btw; add a question after it, and @-mention the side chat later to bring its context back. |
| & | Hand off to a Cloud Agent | Put it at the start of a CLI message; resume the cloud run from web or phone at cursor.com/agents. |
Built-in skills (18)
| Command / Syntax | What it does | Notes |
|---|---|---|
| /automate | Create an automation | Automations can start from a schedule, a Slack message or a GitHub event. |
| /autopilot | Watch a pull request | Acts on review comments, merge conflicts, failing checks and leftover tasks for that PR. |
| /canvas | Build a React canvas | |
| /create-hook | Create a hook | Also registers the new hook in hooks.json for the matching agent lifecycle event. |
| /create-rule | Create a rule | |
| /create-skill | Create a skill | Sets up the skill's folder layout along with its SKILL.md file. |
| /create-subagent | Create a subagent | The result includes a focused role and instructions for delegating work to it. |
| /cursor-blame | Trace AI-written changes | Shows which prompts produced a given AI-authored change. |
| /loop | Repeat a prompt on a timer | Can repeat a skill as well; paired with /goal and no fixed interval, the agent picks when to wake. |
| /migrate-to-skills | Migrate to Agent Skills | Covers eligible slash commands too, not only dynamic rules. |
| /review | Run a code review | Picks a suitable code-review agent; /review-bugbot and /review-security are the focused variants. |
| /review-bugbot | Review with Bugbot | |
| /review-security | Run Security Review | |
| /sdk | Build with the Cursor SDK | |
| /split-to-prs | Split into smaller PRs | |
| /statusline | Set the CLI status line | The footer can show your own command's output together with live token usage. |
| /update-cli-config | Edit CLI config | Writes changes to ~/.cursor/cli-config.json; /config is the interactive editor for the same settings. |
| /update-cursor-settings | Change an editor setting |
Recipes
Ways to chain Cursor slash commands and @-mentions across the editor chat and the CLI, put together from the reference rows above.
Review your branch before opening a pull request
/review with @Branch
In the editor's Agent chat, @Branch attaches every change on your branch compared with main, and /review picks a suitable code-review agent for it. Use @Commit instead when only the uncommitted changes matter.
Plan first, then let the CLI run
/plan [prompt] → review the plan → /run-everything on
In Plan mode the agent asks clarifying questions and designs the approach before any code is written. Once the plan looks right, Run Everything stops the per-command approval prompts; /run-everything status shows the current setting.
Keep a long CLI session usable
/context → /summarize (not /clear)
/context shows what is filling the context window. /summarize compresses the conversation and keeps your latest message word for word, while /clear starts an empty chat. /rewind is the choice when recent turns, and their file changes, should be undone.
Ask a side question while the agent works
/btw in the CLI, /side in the editor
Both leave the main conversation on track. The CLI answer appears in an overlay and never enters the history; the editor's side chat is a conversation of its own that you @-mention later to pull its findings back.
Work across several repositories
/add-dir → /save-workspace → /load-workspace
Add each directory during a CLI session, save the set under a name, and reload it in a later session instead of adding the directories again.
Keep a skill on for a whole task
/skill-name, then Alt+Enter (Option+Enter on Mac)
Enter attaches a skill to one message, and it fades as the conversation moves on. Alt+Enter, or Use as Mode, keeps it in context on every turn until you leave the mode, which suits a review checklist or a team playbook.
Frequently Asked Questions
Do these slash commands work in the Cursor editor chat?
The groups from Models and modes to Help and account are commands of the Cursor CLI, the agent command in a terminal. In the editor's Agent chat, / runs skills, such as the built-in skills at the end of the table, and /side and /goal work there too. The @ mentions are documented for the editor chat; in the CLI, @ attaches files and folders.
What does @Commit attach compared with @Branch?
@Commit attaches the diff of your uncommitted working state (listed as Diff of Working State). @Branch attaches the full diff of your branch against main (listed as Diff with Main), the usual context for a review prompt.
How do I keep a skill active across a whole session?
Pick it from the / menu and press Alt+Enter (Option+Enter on Mac), or choose Use as Mode. The skill then stays in context on every turn until you exit the mode; plain Enter attaches it to one message only. Custom Modes work in the Agents Window and the CLI.
What is the difference between /btw in the CLI and /side in the editor?
In the CLI, /btw asks a quick read-only question while a turn runs and shows the answer in an overlay that never enters the chat history. In the editor chat, /side (or /btw) opens a side chat: a separate, lasting conversation that sees the main thread, which you @-mention later to bring its context back.
How do I move a CLI conversation to a Cloud Agent?
Start a message with &. The conversation goes to a Cloud Agent that keeps running while you are away, and you can continue it on the web or on mobile at cursor.com/agents.
Sources and checks
- Cursor Docs: CLI slash commands reference (checked 2026-10-10) CLI slash-command table: commands, arguments and aliases
- Cursor CLI changelog, January 2026 to release v2026.09.28 CLI commands announced in release notes but missing from the reference table; behaviour notes
- Cursor Docs: Prompting agents @ mentions in the chat input; skills as Custom Modes
- Cursor Docs: Cursor Agent overview Side chats (/side, /btw) and /goal in the chat input
- Cursor Docs: Using Agent in CLI Cloud Agent handoff with &, @ in the CLI, input keys
- Cursor Docs: Agent Skills Built-in Cursor skills table; /skill-name invocation
Every row on this page was matched to a line in the documentation above on . A command or parameter that could not be found there is not listed.
1 row from the earlier version of this page was removed in that check: duplicates, and entries the documentation does not list or lists differently.
AI tools change their commands, parameters and model names between releases: the vendor's current documentation and the version you run have the last word. Found a wrong or missing entry? Tell us and it gets checked against the source. How we verify describes the process and its limits.
Related prompt references
Claude Code (157) · GitHub Copilot (59) · Aider (46) · All AI prompt references