Cursor Commands & Shortcuts

Key points

• Type / after a folder name to expand subfolders.
• Navigate candidates with arrow keys; press Enter to select.

Examples

@src/components/Header.tsx

Key points

• More granular than @Files.
• Directly specify functions or classes.

Examples

@handleSubmit

Key points

• In addition to built-in docs, you can add your own URLs.
• Use the "Add new doc" option to register new documentation URLs.

Examples

@Docs → Next.js

Key points

• Useful when asking about recent changes or review context.
• Lets the model reason about repository history without pasting diffs manually.

Examples

@Git → "Summarize what changed on this branch"

Key points

• Useful for release notes, API references, and third-party documentation.
• Use when internal project files are not enough context.

Examples

@Web https://docs.cursor.com

Key points

• Useful when asking the agent to follow or explain repository-specific rules.
• Surfaces rule context without manually copying rule files into the prompt.

Examples

@Cursor Rules

Key points

• Useful for carrying over decisions and discussions from previous conversations.

Examples

@Past Chats

Key points

• Use when context runs low in long conversations.
• Auto-compression also occurs, but this allows explicit manual execution.

Examples

/summarize

Key points

• Custom command: can be defined in .cursor/commands/commit.md.
• Can be combined with /pr.

Examples

/commit

Key points

• Custom command: can be defined in .cursor/commands/pr.md.
• Can be used with /commit: /commit and /pr

Examples

/commit and /pr these changes to address DX-523

Key points

• Project-level: .cursor/commands/*.md
• Global: ~/.cursor/commands/*.md
• Team: Created via Cursor Dashboard (Team/Enterprise plans).

Examples

.cursor/commands/code-review.md

Key points

• Auto-generates rule files under .cursor/rules/.
• Creates files with frontmatter (alwaysApply, globs, etc.).

Examples

/create-rule

Key points

• Automatically converts rules with 'Apply Intelligently' configuration.
• Preserves original behavior patterns when migrating slash commands.

Examples

/migrate-to-skills

Key points

• Agent works in an independent worktree, keeping your main code safe
• Ideal for parallel branch work and experimental changes

Examples

/worktree

Key points

• Compare outputs from multiple models simultaneously
• Useful for comparing code generation and refactoring quality

Examples

/best-of-n

Key points

• The agent forms hypotheses, adds log statements, and uses runtime info to narrow down the cause
• Applies a targeted fix only after the root cause is identified, avoiding shot-in-the-dark edits
• Works with Auto-run so suggestions can be executed automatically inside Debug Mode

Examples

/debug

Key points

• Ask a side question to the agent without pausing the main task
• Seamlessly return to the original task once the question is answered
• Handy for verifying the rationale behind a recent change
• In the Agents Window, this can also start a side chat — a separate thread that doesn't interrupt the main conversation, just like /side or the plus button (Cursor 3.11).

Examples

/btw btw why did you change this function?

Key points

• List and edit CLI settings directly from within the session
• Switch models, default behaviors, and runtime preferences through a GUI-like panel
• Pairs with the /update-cli-config skill to apply configuration changes automatically

Examples

/config

Key points

• Always show the current mode (Agent / Ask / Debug, etc.) in the status bar
• See branch name, runtime environment, working directory, and worktree at a glance
• Pin session metadata and hints about the active task in the status area

Examples

/statusline

Key points

• Dispatches requests to async subagents running in parallel instead of adding them to the queue
• Breaks larger tasks into smaller chunks for a fleet of subagents to tackle simultaneously
• Can switch already-queued messages over to multitask mode without waiting for the current run to finish

Examples

/multitask

Key points

• Specify a fixed interval to repeat the prompt on that schedule (e.g. every 5 minutes)
• Omit the interval and the agent decides when or what event should trigger the next run
• Use 'until tests pass' or 'until you stop it' as a termination condition for long-running local agent tasks

Examples

/loop 5m check deploy status

/loop work on this feature until tests pass

Key points

• Offload the next submitted task to a cloud subagent running in an isolated VM
• Keep your local session free while the cloud agent works in parallel
• Hand off seamlessly between local and cloud sessions (Cursor 3.7, Cloud Environment Setup)

Examples

/in-cloud

Key points

• Delegate PR preparation to a cloud subagent that iterates remotely
• Trigger it from the quick-action pill or by typing /babysit
• Free up the local session while the cloud agent gets the PR merge-ready

Examples

/babysit

Key points

• Describe an automation in plain language and let Cursor set up triggers, instructions, and tools
• Create automations without leaving your local agent session
• Pair with Slack emoji and GitHub triggers to run automations on events

Examples

/automate

Key points

• Run both Bugbot and Security Review in one command before pushing
• Catch bugs and security issues earlier in your local workflow
• Use /review-bugbot or /review-security to run a single reviewer

Examples

/review

Key points

• Run Bugbot on its own to find likely bugs in your changes
• Faster than the combined /review when you only need bug detection
• Bugbot is now over 3x faster, ~22% cheaper, and finds ~10% more bugs (2026-06-10)

Examples

/review-bugbot

Key points

• Run Security Review on its own to surface potential vulnerabilities
• Use when you only need a security check before pushing
• Complements /review-bugbot for targeted reviews

Examples

/review-security

Key points

• You can also start a side chat with /btw or the plus button at the top of the chat panel.
• Side chats are independent Agent conversations that inherit context from the main chat and can be referenced later via at-mentions.
• Designed for reading, searching, and answering — without derailing the main task.

Examples

/side

Key points

• Requires the Cursor Team Kit plugin from the official Cursor marketplace
• Checks the diff against the main branch to scope the cleanup to what you actually changed
• Prefers minimal, behavior-preserving edits that match the surrounding code style

Examples

/deslop

Key points

• Launch with Cmd I. Best for complex features and refactoring.
• Tools: Semantic Search, file read/write, shell execution, browser, image generation, etc.
• Checkpoints automatically save state before changes.

Examples

Cmd I → "Add authentication"

Key points

• Switch with Cmd . → Ask.
• Best for learning, planning, and understanding code.
• Only search tools available (edit tools disabled).

Examples

Cmd . → Ask → "How does this function work?"

Key points

• Switch via Cmd . → Manual.
• Helpful for review-style conversations and cautious step-by-step changes.
• Reduces autonomous actions and keeps you in the loop.

Examples

Cmd . → Manual → "First outline the change plan only"

Key points

• Open Cmd . to select an existing custom mode or create a new one.
• Useful for review-only, docs-only, or team-specific workflows.
• Each mode can define its own instructions and available tools.
• From `/`, pick a skill and press ⌥⏎ (Mac) or Alt+Enter (Windows), or choose Use as Mode, to pin that skill in the chat as an "always on" custom mode.

Examples

Cmd . → Custom Modes → "Review"

Key points

• Shows ghost text (additions) or diff popups (modifications).
• Tab: accept. Escape: reject. Cmd →: accept one word at a time.
• Jump in file: auto-moves to next edit location after accepting.
• Jump across files: portal window suggests edits in other files.

Examples

Tab → Tab (jump)

Key points

• Select code then Cmd K to enter edit instructions.
• Opt Return: ask a question without editing.
• On empty files, can generate the entire file.

Examples

Cmd K → "Add TypeScript types"

Key points

• Useful for coding style, recurring constraints, and preferred workflows.
• Can be used by agents and automations to stay consistent across sessions.

Examples

Save memory: "Use pnpm in this repository"

Key points

• Best-of-N: send the same prompt to multiple models simultaneously and compare.
• Apply button merges worktree changes back to the main branch.
• Up to 20 worktrees per workspace.

Examples

Launch multiple Agents → Apply best result

Key points

• Auto-saved each time the Agent makes changes.
• "Restore Checkpoint" button to revert to any point.

Examples

Restore Checkpoint

Key points

• Review option appears automatically after Agent finishes.
• Detects code quality, bugs, and security issues.

Examples

Find Issues

Key points

• Finds likely bugs, regressions, and quality issues in PRs.
• Can suggest direct fixes from the review findings.

Examples

Bugbot → Autofix

Key points

• Cmd Return: execute the generated command.
• Escape: accept command into shell (without running).

Examples

Terminal → Cmd K → "Search logs from last 3 days"

Key points

• Launch from cursor.com dashboard or via API.
• Computer Use: browser automation, screenshots, and video recording in isolated VMs.
• Bugbot Autofix: detects issues in PRs and auto-proposes fixes (35%+ merge rate).
• New agent conversation hooks (beforeSubmitPrompt, afterAgentResponse, afterAgentThought, stop, subagentStart, and more) let you observe and control prompts, responses, thinking, subagents, and turn completion (Cursor 3.11).

Examples

cursor.com → Cloud Agents → New Agent

Key points

• Useful when a task takes too long for an interactive foreground session.
• Pairs naturally with cloud execution and automations.

Examples

Start a background agent to investigate flaky tests

Key points

• Triggers: schedule (cron), GitHub PR/push, Slack messages, Linear issues, PagerDuty incidents, webhooks.
• Tools: open PR, comment, request reviewers, send to Slack, MCP, Memories, and more.
• Permissions: Private / Team Visible / Team Owned.

Examples

cursor.com/automations/new

Key points

• Built-in subagents: Explore (code search), Bash (shell execution), Browser (browser control).
• Two modes: foreground (blocks until done) and background (returns immediately).
• Custom subagents can have dedicated prompts, tools, and models.

Examples

Agent → delegate research to subagent

Key points

• Displays partner UIs from Amplitude, Figma, tldraw, and more directly inside Cursor.
• MCP tools can return interactive UI alongside standard tool output.

Examples

MCP Apps → view diagram in tldraw

Key points

• Open via Cmd Shift P → 'Agents Window'
• Can be used alongside the traditional IDE view; switch back anytime.
• Run multiple agents across repositories and environments simultaneously

Examples

Cmd Shift P → Agents Window

Key points

• ⌘+Shift+D: toggle Design Mode
• Shift+drag: select an area
• ⌘+L: add element to chat, ⌥+click: add element to input

Examples

⌘+Shift+D

Key points

• Display multiple chats simultaneously for parallel work
• Switch between side-by-side and grid layouts

Examples

Agent Tabs

Key points

• Split the current view into panes to manage several agents at once
• Drag agents between tiles to rearrange them on the fly
• Use keybindings to quickly navigate between tiles

Examples

Tiled Layout

Key points

• Windows/Linux: Ctrl K
• Search goes beyond names and PR numbers into conversation content.
• Local indexing supports searching thousands of conversations quickly.

Examples

Cmd K → Search transcripts

Key points

• The install command handles dependency installation, code generation, and cache warming ahead of time, while the start command only brings up the fresh services each session needs
• If a deploy or update breaks the environment, agents keep running on the last successful build; failed builds never go live and you get notified so you can debug asynchronously
• The Builds tab in the Cloud Agents dashboard shows build status, logs, commit SHAs, and which build each agent run used, plus manual build triggers and a configurable staleness threshold
• New environments get builds automatically; existing environments enable them from Enable Builds in the Builds tab

Examples

cursor.com → Cloud Agents → Environments → Builds

Key points

• Select Start from scratch in the repo picker, then prompt the agent — no third-party SCM connection is required to get going
• Once the agent's build is to your liking, click Create repo to save the work in an Origin repo: choose a custom or suggested name, set visibility to private or internal, then find the repo under the Codebase tab
• Cursor port-forwards the cloud agent's live environment straight to your browser, so you can preview it and use tools like design mode
• Connect a Vercel account and hit publish to get a live URL for what you built (a Vercel account is required for publishing)

Examples

cursor.com → Cloud Agents → Start from scratch

Key points

• Location: .cursor/rules/*.md or .cursor/rules/*.mdc
• Apply types: Always Apply, Apply Intelligently, specific files only, manual.
• Use frontmatter to set alwaysApply: true or globs patterns.

Examples

.cursor/rules/coding-standards.md

Key points

• Location: Cursor Settings → Rules
• Project rules take precedence (Team > Project > User).

Examples

Cmd Shift J → Rules

Key points

• Write rules in Markdown format.
• Functional alternative to .cursor/rules.
• .cursorrules (legacy) is still supported.

Examples

AGENTS.md

Key points

• Useful for bootstrapping .cursor/rules from the current repository style.
• Can help turn team conventions into reusable Cursor rules.

Examples

Cmd Shift P → Generate Cursor Rules

Key points

• Modes: Run in Sandbox / Ask Every Time / Run Everything
• Network access: sandbox.json only / defaults included / allow all.
• Protections: browser auto-run, file deletion, dotfile, and external file protection.

Examples

Cmd Shift J → Agents → Auto-Run

Key points

• Partial Accepts: accept one word at a time with Cmd →.
• Toggle suggestions in comments, whitespace-only suggestions, auto-import (TypeScript/Python).
• Status bar: snooze, disable globally, or disable for specific file extensions.

Examples

Cmd Shift J → Tab

Key points

• Location: .cursorignore in project root.
• Same pattern syntax as .gitignore.
• Exclude sensitive files or binaries from AI.

Examples

.cursorignore → .env*

Key points

• Location: Cursor Settings → MCP, or .cursor/mcp.json
• Configure MCP tool allowlists for sandbox execution.

Examples

.cursor/mcp.json

Key points

• Location: .cursor/worktrees.json
• Setting: cursor.worktreeCleanupIntervalHours (cleanup interval).
• cursor.worktreeMaxCount: max worktrees per workspace (default 20).

Examples

.cursor/worktrees.json

Key points

• Locations: .agents/skills/, .cursor/skills/, ~/.cursor/skills/
• SKILL.md contains metadata (name, description) and documentation.
• Invoke manually with /skill-name or let the agent use it automatically.

Examples

.cursor/skills/code-review/SKILL.md

Key points

• Locations: .cursor/hooks.json (project), ~/.cursor/hooks.json (user)
• Events: sessionStart/End, preToolUse/postToolUse, beforeShellExecution, and more.
• Two types: command-based (shell scripts) or prompt-based (LLM-evaluated).

Examples

.cursor/hooks.json → preToolUse

Key points

• Manifest: .cursor-plugin/plugin.json (name field required).
• Published on Cursor Marketplace. Team/Enterprise can use private marketplaces.
• All marketplace plugins undergo security review before listing.

Examples

.cursor-plugin/plugin.json

Key points

• Configured from the Cursor dashboard.
• Precedence: Team Rules > Project Rules > User Rules.

Examples

cursor.com → Team Rules

Key points

• Allowlisted calls run immediately; sandboxable calls run in the sandbox
• Remaining actions go through a classifier subagent that decides whether to permit, try alternatives, or ask for approval
• Add custom instructions to steer the classifier subagent's behavior

Examples

Cmd Shift J → Agents → Run Mode

Key points

• On native Windows, run irm 'https://cursor.com/install?win32=true' | iex from PowerShell
• After installing, verify with agent --version and add ~/.local/bin to your PATH
• Auto-update is on by default; update manually with agent update or turn it off with --disable-auto-update
• The August 11, 2026 release made the install script fall back to wget or python3 when curl is broken

Examples

curl https://cursor.com/install -fsS | bash

agent --version

Key points

• Run with no arguments for an interactive session, or pass a prompt to start with an initial instruction
• Supports the same Agent / Plan / Ask modes as the editor, cycled with Shift+Tab
• Add files and folders to context with @, and double-press Ctrl+D to exit
• Automatically loads AGENTS.md, CLAUDE.md, and .cursor/rules and applies them as rules

Examples

agent

agent "refactor the auth module to use JWT tokens"

Key points

• Defaults to All chats; use Left/Right to switch to This workspace and scope it to the current directory
• Chats from other directories show a folder label, and resuming one loads the full conversation
• The August 11, 2026 release scans sessions in parallel, so the picker opens quickly even on network filesystems

Examples

agent ls

Key points

• agent resume picks up the latest conversation, and --continue does the same
• Pass a thread id to --resume to load a specific thread, or open the picker in-session with /resume or Ctrl+Y
• Cancelling /resume keeps whatever you had already typed (June 22, 2026 release)
• Completed subagents persist checkpoints, so resuming restores their prior context

Examples

agent resume

agent --resume="chat-id-here"

Key points

• Check your current version with agent --version
• Run the same update from inside a session with /update (June 9, 2026 release)
• Turn background auto-updates off with --disable-auto-update
• Since the July 13, 2026 release, install directories in use by a running session are skipped by auto-update cleanup

Examples

agent update

agent --version

Key points

• Press q during login to reveal a QR code for the same login URL; narrow terminals and non-interactive sessions show the URL only
• For scripts, set your API key in the CURSOR_API_KEY environment variable instead
• In sandboxes without the macOS Keychain, set AGENT_CLI_CREDENTIAL_STORE=file to store credentials in an owner-only file
• The August 11, 2026 release sends and enforces MDM sign-in policy, including organization, team, email, and domain allowlists

Examples

agent login

export CURSOR_API_KEY=your_api_key_here

Key points

• Reads the same connection state that /mcp shows inside a session
• The June 29, 2026 release fixed working servers being shown as disconnected when their tools or instructions were available
• Since the July 13, 2026 release, a server configured in more than one scope runs as a single instance instead of duplicates

Examples

agent mcp list

Key points

• add registers a marketplace by git URL; pin a branch, tag, or commit with --git-ref
• list prints each marketplace's name, scope, and git URL, with --format json for scripts
• update re-indexes one from its repository, and remove deletes a user-scoped marketplace

Examples

agent plugin marketplace add https://github.com/acme/cursor-plugins

agent plugin marketplace list --format json

Key points

• Supports both access keys and a team IAM role
• Works across interactive, headless, and editor sessions
• The March 2026 release added Claude Sonnet 4.6 to the bring-your-own-key model list

Examples

agent bedrock

Key points

• --pool joins a named pool so you can split fleets by purpose
• Workers release after an hour idle by default; --idle-release-timeout 0 keeps them running
• Repeat --worker-dir to span multiple repositories, and start from a non-git directory
• A data-directory lock prevents duplicate daemons, and session-start and session-end hooks run on claim and release
• Workers started with --computer-use --share-desktop share a managed desktop instead of a dedicated session (August 26, 2026 release)
• agent worker controller can wake claimed, hibernated workers on follow-up through the --spawn hook
• queue watch moved from polling to server-sent events (SSE), so it holds up better against rate limits

Examples

agent worker --pool build-fleet

agent worker --idle-release-timeout 0

Key points

• ACP clients such as Zed and JetBrains can select models and switch modes
• Host-provided MCP servers are respected, with richer streaming that includes thinking and file locations
• Session resume, custom slash commands, and skill invocation all work through the editor

Examples

agent acp

Key points

• The short form is -p; only the response is printed, so it pipes cleanly into other tools
• By default changes are proposed but not applied — add --force (or --yolo) to actually modify files
• Combine with --output-format to choose text, json, or stream-json
• The August 11, 2026 release detects and retries wedged uploads on long-running headless sessions

Examples

agent -p "What does this codebase do?"

agent -p --force "Add JSDoc comments across src"

Key points

• text prints the final answer as plain text (the default for --print)
• json emits structured output that is easy to parse in scripts
• stream-json tracks progress message by message; add --stream-partial-output to stream deltas incrementally
• Per-turn input, output, and cache token totals plus a request_id are included, so you can measure cost

Examples

agent -p --output-format json "Review the recent changes"

agent -p --output-format stream-json --stream-partial-output "Summarize the project structure"

Key points

• Model variants such as fast or high-effort slugs keep their parameters in headless mode
• Since the August 11, 2026 release, a headless selection that requires Max Mode is sent as-is instead of being clamped down
• Inside a session, switch with /model or shortcuts like /opus and /composer
• Since the July 6, 2026 release, fresh installs default to Auto routing

Examples

agent -p "find and fix performance issues" --model "gpt-5"

Key points

• --mode=plan can also be written --plan; it designs an approach before writing code
• --mode=ask explores the codebase read-only without editing files
• Inside a session, switch with /plan and /ask, or cycle modes with Shift+Tab
• The August 11, 2026 release moved mode-switch approvals to a pager UI with a 15-second countdown that auto-rejects when unattended

Examples

agent --mode=plan "Plan the migration of the checkout flow"

agent --mode=ask "Where is authentication implemented?"

Key points

• --yolo is the fully autonomous alias that auto-approves trust, MCP, and command approvals together
• Without --print, headless changes stay proposals instead of being applied
• Inside a session, toggle Run Everything with /auto-run, which is independent of the Shift+Tab mode cycle
• Allowlists live in permissions.json, the same file the editor reads

Examples

agent -p --force "Refactor this code to use modern ES6+ syntax"

Key points

• Previously headless-only, it now settles trust up front in interactive sessions as well
• It records the same saved trust decision the dialog writes when you accept
• Non-interactive runs in untrusted workspaces fail with guidance unless --trust (or --force) is passed
• Setup commands from .cursor/worktrees.json run only once the workspace is trusted (July 13, 2026 release)

Examples

agent --trust "run the tests and fix the failures"

Key points

• Worktrees are created under ~/.cursor/worktrees by repository and worktree name, alongside worktrees created from the editor
• Omit the name and Cursor generates one; pick the base branch with --worktree-base
• Combine with --workspace when you need an explicit repository root
• CLI worktrees are cleaned up with the same retention rules the editor uses

Examples

agent --worktree "upgrade the test runner and fix any broken snapshots"

agent --workspace ~/src/my-app --worktree auth-fix "fix the flaky auth test and open a PR"

Key points

• Combine with --workspace to pin an explicit repository root while adding extra directories
• To add directories mid-session use /add-dir, which refreshes slash skills and custom commands immediately
• Restart only when you want the agent to discover new skills automatically

Examples

agent --add-dir ../api --add-dir ../web "align the shared type definitions"

Key points

• Load a plugin you are still developing to test it before publishing
• The August 11, 2026 release made hooks defined by plugins loaded with --plugin-dir execute and refresh on reload
• Local plugins can also be pointed at from the enabled_plugins key in ~/.cursor/settings.json

Examples

agent --plugin-dir ./my-plugin

Key points

• Inside a session, /sandbox opens an interactive menu for sandbox mode and network access, and the settings persist across sessions
• Sandbox policy lives in ~/.cursor/sandbox.json and .cursor/sandbox.json, and is honored over SSH
• If sandboxing is enabled but unavailable, the CLI fails fast with a clear reason
• The August 11, 2026 release added a read-access boundary to the policy and routes sandboxed git-over-SSH through the SOCKS proxy

Examples

agent --sandbox enabled

Key points

• Shell, MCP, and Fetch calls are checked in order: allowlisted calls run immediately, sandboxable calls run in the sandbox, and the rest go to a classifier
• Inside a session, toggle it with /auto-review or from /config
• Steer the classifier with allow and block instructions in permissions.json
• Since the June 29, 2026 release the classifier can also inspect invoked skill and file-backed custom-command files; availability is admin-controlled

Examples

agent --auto-review "update the dependencies and get the tests passing"

Key points

• The short form is -H; pass it multiple times for multiple headers
• Useful for auth headers required by a corporate proxy or gateway
• For the proxy itself, the HTTPS_PROXY environment variable is honored for both auth and agent streams

Examples

agent -H "X-Team: platform" -H "X-Env: staging" -p "review the changes"

Key points

• Shows an active or paused status line so the current goal stays visible
• Pressing Ctrl+C pauses the goal
• It is rolling out and gated, so it may not appear in every environment yet

Examples

/goal

Key points

• Shows included-usage meters with Auto and API breakdowns
• Displays on-demand spend against your limit, your plan name, and the billing-cycle reset date
• Enterprise pooled plans get a spend chart comparing the current billing cycle with the previous period of the same length, plus a dashboard link

Examples

/usage

Key points

• Type something like /model sonnet and the picker opens pre-filtered without submitting your prompt
• Shortcuts such as /opus and /composer jump straight to a model, and each family remembers your last choice
• /fast toggles Fast when the current model supports it; disable shortcuts under Model slash commands in /config
• The August 11, 2026 release pins Auto to the top of the list and selects a model when you edit its parameters; set Max Mode explicitly with /max-mode

Examples

/model

/model sonnet

Key points

• An interactive timeline restores files and conversation state to any earlier turn
• Includes per-turn file diffs and a conversation-only restore; /undo and /restore are aliases
• Rewinding also restores that turn's image attachments to the prompt (May 20, 2026 release)

Examples

/rewind

Key points

• Duplicates the conversation while keeping the original, so you can try another approach in parallel
• /branch and /duplicate are aliases
• Forked conversations can be resumed individually from agent ls or the /resume picker

Examples

/fork

Key points

• Shows the breakdown across rules, skills, MCPs, and subagents
• When context runs tight, use it to find what to trim before compacting with /summarize
• The footer also keeps working directory, git branch, and context information visible

Examples

/context

Key points

• Browse tool schemas, log in, log out (clearing stored OAuth credentials), enable, and disable without leaving the session
• Servers are grouped under User / Project / Team scope, and failed servers show their actual error
• When the same server exists in multiple scopes, the Configure Scope tab picks which configuration wins, saved per project (July 13, 2026 release)
• The August 11, 2026 release logs into known hosts like Slack with default authentication, avoiding dynamic-client-registration failures

Examples

/mcp

Key points

• Add a marketplace by passing a git URL to /plugin marketplace add; browsing and managing by scope is supported
• A plugin's skills, slash commands, subagents, and MCP servers all load into the session together
• /plugins lists them, and each plugin's MCP servers link into /mcp management
• Reloading refreshes slash commands and the palette, and MCP tools survive the reload

Examples

/plugin

/plugin marketplace add https://github.com/acme/cursor-plugins

Key points

• The Session tab narrows the diff to only the edits made in this session
• Press i to add follow-up instructions, Up/Down to scroll, and Left/Right to switch files
• Press o or O to open diffs in your editor, and use the zen toggle while reviewing
• File edits render borderless with memoized diffs, so large edits do not slow the UI (June 9, 2026 release)

Examples

/changes

Key points

• Invoke a skill directly with /skill-name; this works in --print headless runs too
• /rules and /commands browse and inline-edit rules and custom commands
• Skills load from .cursor/skills plus .claude/skills, .agents/skills, and .codex/skills, including nested subdirectories and symlinked directories
• The August 11, 2026 release attaches a skill slash entry once on Enter, and Option+Enter invokes a mode-backed skill as a sticky custom mode

Examples

/skills

Key points

• Debug logs are written for every session, so you can attach them directly to a bug report
• On shared machines logs are written per user to avoid permission errors (June 22, 2026 release)
• The path lingers on screen longer, so you can still read it if the copy is missed

Examples

/logs

Key points

• Pick one step of a multi-step reply from a per-step picker
• Copy the agent's responses alongside your own messages, and copy long replies without the terminal stalling
• Related: /rename names a session, and ? opens the input shortcut cheat sheet

Examples

/copy

Key points

• Visual-mode selection with delete and change operators, and the active Vim mode shows in the footer
• Find motions f, F, t, T, semicolon, and comma work with operator composition such as df, and ct.
• Ctrl+G opens the prompt in $EDITOR so you can compose in your real editor and drop the result back
• While revising a plan, Esc returns to Vim normal mode

Examples

/vim

Key points

• Add directories mid-session with /add-dir, then save the set with /save-workspace
• Reload a saved set later with /load-workspace
• To start scoped, use --workspace at launch, and --add-dir to pass extra directories

Examples

/save-workspace

/load-workspace

Key points

• Pick the handed-off task back up on web or mobile at cursor.com/agents
• For Git repositories with Cloud Agents access, the selected model and workspace path are preserved (June 29, 2026 release)
• Transfer status is shown, Esc or Ctrl-C cancels, and failure details remain visible

Examples

& refactor the auth module and add comprehensive tests

Key points

• Open the interactive settings editor with /config instead of hand-editing JSON (April 2026 release)
• The built-in /update-cli-config skill can apply configuration changes for you
• Point statusLine at your own command to render its output in the prompt footer, and choose the Explore subagent's model from /config
• Since the July 6, 2026 release each write stages to a temp file before an atomic rename, so concurrent CLI processes cannot corrupt it

Examples

/config

Key points

• Toggle it under Permissions in /config, or set autoAcceptWebSearch directly in cli-config.json
• Honored in interactive, headless, subagent, and Agent Client Protocol runs
• Approval policy for shell, MCP, and Fetch calls is handled separately by --auto-review and permissions.json

Examples

/config → Permissions → Auto-Accept Web Search

"autoAcceptWebSearch": true

Key points

• Reconnect to a running persistent session with agent persist attach
• List persistent sessions with agent persist list and stop one with agent persist stop
• agent persist --resume continues an existing chat as a persistent session
• To step away without stopping the agent, use /detach — the agent keeps working

Examples

agent persist

agent persist attach

Key points

• Use it in a persistent session started with agent persist — the agent carries on working after you leave
• Come back to the same session afterwards with agent persist attach
• To end the session itself, use agent persist stop instead of /detach

Examples

/detach

Key points

• After installing, verify with origin --version
• If your shell reports command not found: origin, add ~/.local/bin to your PATH (~/.zshrc for zsh, ~/.bashrc for bash)
• Sign in with origin auth login, which also configures the git credential helper so git push and git pull against Origin remotes just work
• Update with origin update; the default channel is stable and origin config set-channel latest|stable switches it

Examples

curl -fsSL https://downloads.cursor.com/origin/install.sh | sh

origin --version

Key points

• login --api-key <key> signs in with an API key instead of the browser flow; when CURSOR_API_KEY is set it is used without the flag
• login --local writes the credential helper to this repository's git config instead of the global one
• auth setup-git reconfigures the credential helper (--host / --global / --local)
• auth status shows the current authentication method and account; auth logout clears stored credentials

Examples

origin auth login

origin auth status

Key points

• repo create-mirrored <github-repo> creates a mirror of a GitHub repository (--namespace / --github-enterprise-id)
• repo list shows the repositories your account can access (--namespace limits it to one namespace)
• repo view [repo] displays a repository and can emit JSON with --json org,name,defaultBranch
• repo clone <repo> [directory] clones over HTTPS using your saved login
• repo delete <repo> prompts for confirmation; -y / --yes is required in a non-interactive shell
• create --default-branch <branch> sets the default branch (server default: main)

Examples

origin repo create my-project

origin repo clone acme/checkout

origin repo create-mirrored acme/checkout

Key points

• create takes -t/--title, -b/--body, -F/--body-file, -H/--head, -B/--base, --push, and -f/--fill (the default status is draft)
• list and status show pull requests in a repo and the ones relevant to you (-s/--state, --mine, --file, -L/--limit, --json, -q/--jq)
• view [target] switches output with --diff, --checks, -c/--comments, and -w/--web; diff and checks are also standalone subcommands
• checkout [target] fetches the pull request branch, edit changes title/body/base, and ready marks a draft as ready
• review --approve files a review, comment -b adds a discussion comment, and thread list/resolve/reopen/reply drives review threads
• merge, close, and reopen handle the lifecycle; merge --auto merges once the requirements are met
• -R/--repo org/name targets a repository other than the one behind your current origin remote

Examples

origin pr create -t "Fix CI"

origin pr list --mine

origin pr review --approve

Key points

• ruleset list shows the rulesets configured for a repository
• ruleset view <ruleset-id> shows one ruleset; the ID comes from the list output
• -R/--repo org/name targets a specific repository
• The CLI is read-only here — create and edit rulesets from repository settings on the web

Examples

origin ruleset list

origin rs list

Key points

• ssh-key add [key-file] registers a public key; -t names the key
• ssh-key list shows the keys on your account (--json for JSON output)
• ssh-key delete <id> removes a key by the ID printed by ssh-key list

Examples

origin ssh-key add ~/.ssh/id_ed25519.pub

origin ssh-key list

Key points

• Supports -X/--method, -H/--header, -F/--field, -f/--raw-field, --input, and -q/--jq
• The owner, repo, and branch placeholders in a path expand from -R/--repo, the ORIGIN_REPO environment variable, or the origin git remote
• --auth-token (CURSOR_AUTH_TOKEN) passes a bearer token directly
• --endpoint (CURSOR_ORIGIN_ENDPOINT) changes the API server; the default is https://origin.cursor.com

Examples

origin api /repos/acme/checkout

origin api /repos/acme/checkout -q .defaultBranch

Key points

• Enable it by appending to your shell config, for example origin completion >> ~/.zshrc
• With completion installed, branch-valued flags such as --head and --base complete local branch names
• Every command takes --help (origin --help / origin repo --help / origin pr create --help)

Examples

origin completion >> ~/.zshrc

origin --help

Key points

• config get-channel shows the current release channel
• config set-channel latest|stable switches channels; the default is stable
• Run origin update after switching to pull the build from the new channel

Examples

origin config get-channel

origin config set-channel latest

Key points

• The default channel is stable; origin config set-channel latest switches to early releases
• Verify the result with origin --version after updating
• The Cursor Agent CLI (agent) is a separate binary and is updated with agent update

Examples

origin update