Cursor cheatsheet

Lookup, not a tutorial. How-to: tutorial. Definitions: glossary.

Contents

• Decision tables
• Capability matrix
• Term index
• Keyboard shortcuts
• Config
• Common errors
• Templates
• Sources

---

Decision tables

Which surface

Job	Use	Why
Finish the line / next edit	Tab	Cheap, no agent turn
This selection or function	Inline Edit Cmd+K	Selection is the context
Questions only	Ask	Harder to clobber files
Multi-file work, commands	Agent	Tool loop
Fuzzy, many files	Plan	Align first
Reproducible, cause unknown	Debug	Runtime evidence
PR review	Bugbot	Diff, not your running app
Away / parallel / isolated VM + PR	Cloud Agents	Formerly Background Agents
Terminal or headless CI	CLI agent	Binary is agent, not cursor
Point / draw / voice on the running UI	Design Mode	Agents Window browser; Cmd+Shift+D
Host / mirror / browse the repo	Origin	Git forge; binary origin
PR security gate or cron scan	Security Agents	/review-security; team usage pool
Assign reviewers / approve low-risk PRs	PR Routing	Exact file APPROVAL_POLICY.md
Same agent inside your process	SDK	@cursor/sdk / cursor-sdk / Bridge

Rules vs Skills vs Commands vs Hooks vs MCP vs Subagents

Job	Pick	Why
Short constraint on every turn	Rules	In the system prompt
Human-readable brief for several agents	AGENTS.md	No frontmatter
On-demand multi-step workflow	Skills	Not always in context
Prompt you type every day	Commands	Explicit /name
Script before/after a tool	Hooks	Not the model's memory
External system	MCP	Tools protocol
Isolated context or parallel work	Subagents	Own window

New chat vs continue

Situation	Pick
New feature / module	New chat
Agent looping on the same mistake	New chat + @Chats
Next iteration of the same feature	Continue
Debugging what it just wrote	Continue

---

Capability matrix

Moved from the Chinese overview (docs/zh/products/ai-coding/index.md). URLs updated to pages that resolved in August 2026. Bugbot's description is corrected to match official docs.

Feature	What it is	Official docs
AGENTS.md	Plain-markdown project instructions	Rules · AGENTS.md
Rules (.cursor/rules)	.mdc + glob / intelligent / manual / Always	Rules
Commands	.cursor/commands/*.md, / trigger	Commands
Skills	SKILL.md packs; /migrate-to-skills	Skills
MCP	External tools and data	MCP
Hooks	Agent / Tab / workspace lifecycle scripts	Hooks
Sub-agents	Isolated delegates; built-in Explore / Bash / Browser	Subagents
Bugbot	PR review (bugs / security / quality); Autofix spawns Cloud	Bugbot
Cloud Agents	Isolated VMs, PRs; formerly Background Agents	Cloud Agent
Cursor CLI	Terminal agent; headless agent -p	CLI · Installation
Design Mode	Visual prompts in the Agents Window browser	Design Mode
Origin	Cursor git forge (early beta); Pro / Teams / Enterprise	Origin
Security Agents	Security Reviewer (PR) + Vulnerability Scanner (cron)	Security Agents
PR Routing & Approval	Route reviewers; approve low-risk PRs	Approval
TypeScript SDK	@cursor/sdk; local or cloud runtime	TS · Python · Bridge
Modes	Agent / Ask / Plan / Debug. Cmd+.; Plan via Shift+Tab	Overview · Plan · Debug
Tab	Multi-line complete, jumps, TS/Python auto-import	Tab
Chat	Agent thread; @ attaches context	Prompting
Codebase Indexing	Vector index; semantic search after ~80%	Semantic search
@ Symbols	@ files / folders / terminals / chats / git / browser	Prompting
Notepad	Legacy matrix row. Not listed on the 2026-08 Prompting page. Status unverified	old mentions URL
Docs Integration	Custom docs in context (legacy mentions capability)	Prompting
Privacy Mode	Code is not used for training	Privacy
Model Selection	Cursor Models pool vs third-party pool	Models & Pricing
Integrations	GitHub / GitLab / Linear / Slack	GitHub
Inline Edit	Cmd+K on a selection	Inline Edit
Reuse Existing Code	Legacy matrix row. Not in the 2026-08 nav. Status unverified	reuse-existing-code
Long-running Agents	Official essay	Blog
Self-driving Codebases	Official essay	Blog

Used daily, not in the original matrix:

Feature	What it is	Official docs
.cursorignore	Blocks index, Tab, Agent, Inline Edit, @	Ignore files
.cursorindexingignore	Index only; tools can still read	same
Keyboard shortcuts	Full table	Keyboard Shortcuts

---

Term index

One sentence each. Longer definitions: glossary.

Term	One line	More
Rules	Persistent instructions at the start of context	Glossary
AGENTS.md	Project brief without metadata	Glossary
Skills	On-demand SKILL.md packs	Glossary
Commands	/ markdown workflows	Glossary
MCP	External tools protocol	Glossary
Hooks	Lifecycle scripts	Glossary
Subagents	Child agents, own context	Glossary
Modes	Agent / Ask / Plan / Debug	Glossary
Tab	Completion model	Glossary
Bugbot	PR review	Glossary
Checkpoints	Snapshots before big edits	Glossary
Codebase Index	Vectors for semantic search	Glossary
Privacy Mode	No training on your code	Glossary
Cloud Agents	Agents on remote VMs; formerly Background Agents	Glossary
CLI agent	Official terminal entry; -p for headless	Glossary
Design Mode	Click / draw / voice in the Agents Window browser	Glossary
Origin	Cursor git forge; claim a codebase name	Glossary
Security Agents	PR reviewer + cron scanner; team pool	Glossary
PR Routing	Assign reviewers; maybe auto-approve	Glossary
SDK	Same agent from @cursor/sdk / cursor-sdk	Glossary

---

Keyboard shortcuts

From the official table. Full list: Cmd+R then Cmd+S (Windows/Linux: Ctrl+R then Ctrl+S), or Command Palette → Keyboard Shortcuts. Everything is remappable.

General

macOS / Windows·Linux	Action
Cmd+I / Ctrl+I	Toggle sidepanel (unless bound to a mode)
Cmd+L / Ctrl+L	Toggle sidepanel; selection → new chat
Cmd+E / Ctrl+E	Toggle Agent layout
Cmd+. / Ctrl+.	Mode menu
Cmd+/ / Ctrl+/	Cycle models
Cmd+Shift+J / Ctrl+Shift+J	Cursor settings
Cmd+Shift+Space / Ctrl+Shift+Space	Voice
Cmd+, / Ctrl+,	General settings
Cmd+Shift+P / Ctrl+Shift+P	Command palette

Chat / Agent

Shortcut	Action
Enter	Default nudge; queues while Agent is working
Ctrl+Enter	Queue
Cmd+Enter / Ctrl+Enter while typing	Force send / inject now
Cmd+Shift+Backspace	Cancel
Shift+Tab in the input	Rotate Agent modes (including Plan)
Cmd+N / Cmd+R	New chat
Cmd+T	New chat tab
Cmd+[ / Cmd+]	Prev / next chat
Esc	Unfocus

Inline Edit / Tab / terminal

Shortcut	Action
Cmd+K / Ctrl+K	Inline Edit
Opt+Enter / Alt+Enter	Quick Question
Tab	Accept Tab suggestion
Cmd+→ / Ctrl+→	Accept next word
Cmd+K in the terminal	Terminal prompt bar

The old stub called Cmd+Shift+K “inline chat” and treated Cmd+L only as “open AI chat”. This table wins.

---

Config

Where files live

Purpose	Project (git)	User
Rules	.cursor/rules/*.mdc	Customize → Rules
AGENTS.md	Root or subdirectory	—
Commands	.cursor/commands/*.md	~/.cursor/commands/
Skills	.cursor/skills/<name>/SKILL.md or .agents/skills/	~/.cursor/skills/, ~/.agents/skills/
Subagents	.cursor/agents/	~/.cursor/agents/
MCP	.cursor/mcp.json	Customize / user MCP
Hooks	.cursor/hooks.json	~/.cursor/hooks.json
Bugbot notes	.cursor/BUGBOT.md	Dashboard team rules
Ignore	.cursorignore, .cursorindexingignore	Global ignore in settings
Plans (saved to workspace)	.cursor/plans/	Home directory by default
Cloud environment	.cursor/environment.json	Dashboard Environments / Secrets
CLI config	.cursor/cli.json	~/.cursor/cli-config.json

Skills / subagents also load .claude/ and .codex/ trees. .cursor/ wins on name clash.

Rules frontmatter

alwaysApply	description	globs	Behavior
true	—	—	Every session. Ignores the other two
false	—	set	Auto-attach when a matching file is in context
false	set	omitted	Agent decides from the description
false	omitted	omitted	Only when @-mentioned

Precedence: Team → Project → User. Merged; earlier source wins conflicts.

Ignore files

.cursorignore uses gitignore syntax. It blocks semantic search, Tab / Agent / Inline Edit reads, and @ mentions.

Official caveat: the terminal and MCP are not bound by .cursorignore. Ignore is not a crypto guarantee.

.env
.env.*
**/*.pem
**/secrets.json
dist/

Index-only exclusion: .cursorindexingignore.

MCP stdio fields

Field	Required	Meaning
type	yes	"stdio"
command	yes	On PATH or an absolute path
args	no	Argument array
env	no	Environment
envFile	no	Extra env file

Interpolation: ${env:NAME}, ${userHome}, ${workspaceFolder}, ${workspaceFolderBasename}, ${pathSeparator} / ${/}.

Remote: url plus optional auth (CLIENT_ID / CLIENT_SECRET / scopes). Desktop redirect http://localhost:8787/callback. Web / Cloud Agents: https://www.cursor.com/agents/mcp/oauth/callback.

Hooks

Priority: Enterprise → Team → Project → User. All matching hooks run; conflicts merge toward the higher source.

Agent events: sessionStart / sessionEnd, preToolUse / postToolUse / postToolUseFailure, subagentStart / subagentStop, beforeShellExecution / afterShellExecution, beforeMCPExecution / afterMCPExecution, beforeReadFile / afterFileEdit, beforeSubmitPrompt, preCompact, stop, afterAgentResponse / afterAgentThought.

Tab: beforeTabFileRead / afterTabFileEdit. App: workspaceOpen.

Exit codes: 0 ok; 2 block (permission: "deny"); anything else fail-open unless failClosed: true.

Cloud agents: command-based project hooks only.

Skills frontmatter

Field	Required	Meaning
name	yes	[a-z0-9-]+, must match the parent folder
description	yes	Relevance hint
paths	no	Globs that gate surfacing
disable-model-invocation	no	true → /name only
metadata	no	Arbitrary map

CLI agent

Sources: Installation, Overview, Headless, Using.

Command	Meaning
curl https://cursor.com/install -fsS | bash	macOS / Linux / WSL install
irm 'https://cursor.com/install?win32=true' | iex	Windows PowerShell install
agent --version	Verify
agent / agent "…"	Interactive
agent --mode=ask	Read-only
agent --mode=plan / --plan	Plan
agent -p "…"	Headless; does not write files by default
agent -p --force "…"	Headless and apply edits (--yolo alias)
& … in a session	Hand off to a Cloud Agent
agent ls / agent resume / agent --continue	Resume
agent update	Manual update
agent --worktree "…"	Edit in a separate git worktree

Script auth: CURSOR_API_KEY. Config: global ~/.cursor/cli-config.json, project .cursor/cli.json (Configuration).

Models and plans (summary)

Authoritative numbers: Models & Pricing (dynamic table).

• Two pools: Cursor Models (Grok 4.6 / 4.5, Composer 2.5) and Other Models (third-party API rates)
• Individual (pre-tax unless noted): Start (India only, ₹649/mo), Pro $20, Pro Plus $60, Ultra $200
• Bugbot and Cloud Agents: Pro and up (Start excludes Bugbot)
• Teams: Standard $40 / user / mo, Premium $120 / user / mo

Per-million token cells were not recoverable from the rendered docs page.

---

Common errors

Symptom	Cause	Fix
.cursor/rules/foo.md does nothing	Rules ignore .md without frontmatter	Rename to .mdc or use AGENTS.md
Always rules blow the window	Too long / too many	Split; keep each under 500 lines; prefer globs
Cannot @ a file	.cursorignore / .gitignore	git check-ignore -v
Empty semantic search	Index below ~80%	Settings → Indexing; check network
Plan still writes code	Reported in the forum	Stop → Restore → “no edits until APPROVE”
Bugbot check is green, comments remain	Default neutral	Read comments; enable fail-on-unresolved to block
Autofix missing	No on-demand usage or Legacy Privacy Mode	Official requirements
Hook missing in Cloud	User-level or type: prompt	Project .cursor/hooks.json, command hook
agent: command not found	~/.local/bin not on PATH	Official Installation PATH step
agent -p did not edit files	Headless proposes only	Add --force / --yolo
Cloud run never starts	SCM not connected or free plan	Admin connects SCM; official Troubleshooting
MCP will not start	command not on PATH / missing env	Customize logs; interpolate secrets
Tab pollutes markdown	Extension not disabled	Status bar → disable for that extension

---

Templates

Project AGENTS.md

# Agent notes

- Package manager: pnpm. Never npm / yarn.
- Checks: `pnpm typecheck`, `pnpm test`.
- Canonical component: `src/components/Button.tsx`.

hooks.json (official stop-loop skeleton)

{
  "version": 1,
  "hooks": {
    "stop": [
      { "command": "bun run .cursor/hooks/grind.ts" }
    ]
  }
}

grind.ts reads JSON from stdin and writes JSON to stdout. Official blog fields: conversation_id, status (completed | aborted | error), loop_count. Return { "followup_message": "..." } to continue. Default loop_limit is 5.

.cursor/BUGBOT.md

# Bugbot

- Amounts are integer cents.
- Copy issues are not bugs unless they change control flow.

---

Sources

Last systematic check: 2026-08-18. Discovery method: repo sources/_template.md.

Official docs

• Cursor Docs — main tree. Child pages listed in references/cursor.md
  • How to read: web-reader, one URL at a time
  • Last verified: 2026-08-18
• Cloud Agents, CLI, Bugbot — three first-class surfaces
  • Last verified: 2026-08-18
• Changelog — editor releases
  • Last verified: 2026-08-18
• Downloads
  • Last verified: 2026-08-18

Chinese locale: prefix /cn/, e.g. cn/docs/rules.

Official long-form

• Best practices for coding with agents
  • Last verified: 2026-08-18
• Long-running agents, Self-driving codebases
  • Last verified: 2026-08-18
• No official cookbook repo analogous to anthropics/claude-cookbooks. Examples live in docs + blog.

Official / core social

• @cursor_ai — product account (footer on cursor.com → X)
  • Layer: product
  • Last verified: 2026-08-18

Company / founder personal accounts are not listed until an official page or staff post names them.

Team blog

• cursor.com/blog — Last verified: 2026-08-18

GitHub

• There is no official public source repo for the Cursor IDE. Do not label third-party mirrors as official.
• GLips/Figma-Context-MCP — community Figma MCP cited by the old stub. Install from that README.
  • Last verified: 2026-08-18 (repo exists)

Aggregators

• cursor.directory — community rules / MCP. Not Anysphere.
  • Last verified: 2026-08-18

Third-party / community

• forum.cursor.com — official forum
  • Last verified: 2026-08-18
• WorkOS on Bugbot
  • Last verified: 2026-08-18

Unverified

• Whether Notepad is still a first-class feature
• Whether https://cursor.com/docs/reuse-existing-code still resolves
• Founder / engineer X accounts with first-party proof
• Whether @modelcontextprotocol/server-filesystem is still the example package to name in docs (JSON above is a field demo)