Rules & AGENTS.md

Language models keep no memory between requests. Every new Agent conversation in Cursor starts from zero unless you give it durable instructions. Rules are Cursor's mechanism for that: Markdown files (and dashboard-managed text) that Cursor inserts at the start of the model context so Agent follows your conventions, commands, and guardrails without being told each time.

Cursor has four rule types and two ways to write them on disk. This page covers all of them, how they load, and what belongs in each.

All persistent-instruction mechanisms in Cursor

Mechanism	Location	Scope	Committed to git?	Managed by
Team Rules	Cursor dashboard (Team and Enterprise plans)	Everyone in the team, all repositories	N/A (stored on Cursor servers)	Team admins
Project Rules	.cursor/rules/*.mdc	Everyone working in the repo	Yes	You
AGENTS.md	./AGENTS.md and any subdirectory	Everyone working in the repo (nested files scope to their directory)	Yes	You
CLAUDE.md	./CLAUDE.md	Everyone working in the repo, always applied	Yes	You
User Rules	Cursor Settings > Rules (Customize > Rules)	You, all projects, synced with your account	N/A	You
User rule files	~/.cursor/rules (%USERPROFILE%\.cursor\rules on Windows)	You, all projects, this machine only (no sync)	N/A	You
Legacy .cursorrules	./.cursorrules	Everyone working in the repo	Yes	You (deprecated)

Rules apply to Agent (Chat) only. They do not affect Tab completion, Inline Edit (Cmd+K), or Bugbot PR reviews. Within Agent they apply in every mode: Agent, Ask, Plan, and Debug.

Project rules (.cursor/rules/)

Project rules are .mdc files in .cursor/rules/. Each file is Markdown with a YAML frontmatter block that controls when the rule is included. The extension matters: a plain .md file in .cursor/rules/ is ignored because it has no frontmatter. If you want plain Markdown, use AGENTS.md instead.

.cursor/rules/
  react-patterns.mdc       # Recognized as a project rule
  api-guidelines.md        # Ignored (wrong extension)
  frontend/                # Subfolders work, but a flat layout is simpler
    components.mdc

Cursor identifies rules by their full path, so two files with the same name in different folders both apply if their conditions match. There is no override by filename.

Frontmatter fields and rule types

Three frontmatter fields decide when a rule is loaded:

Field	Type	Meaning
alwaysApply	boolean	true includes the rule in every conversation and ignores the other two fields
globs	string	Comma-separated file patterns; the rule attaches when a matching file is in context
description	string	Shown to Agent so it can decide whether to pull the rule in

The rule editor exposes these as four rule types. Under the hood they are just combinations of the fields:

Rule type (editor label)	alwaysApply	description	globs	Behavior
Always Apply	true	ignored	ignored	Included in every chat session
Apply to Specific Files	false	optional	provided	Auto-attached when a matching file is in context
Apply Intelligently	false	provided	omitted	Agent reads the description and pulls the rule in when relevant
Apply Manually	false	omitted	omitted	Included only when you @-mention the rule in chat

Older Cursor documentation and community material call these Always, Auto Attached, Agent Requested, and Manual. The behavior is the same.

Always applied:

---
alwaysApply: true
---

- All source files must include the company copyright header
- Never modify generated files in the `dist/` or `build/` directories
- When unsure about implementation details, read the relevant source files before proposing changes

Auto-attached by file pattern:

---
globs: src/components/**/*.tsx
alwaysApply: false
---

- Use named exports, not default exports
- Co-locate styles in a module CSS file next to the component
- Keep components under 200 lines; extract subcomponents into the same directory

Agent-selected based on description:

---
description: RPC service conventions and patterns for the backend
alwaysApply: false
---

- Define each service in its own file under `src/services/`
- Validate inputs at the service boundary before passing data to internal functions
- Return structured error objects with `code` and `message`, never throw raw strings

Manual (only via @-mention):

---
alwaysApply: false
---

- Every database migration must have both `up` and `down` functions
- Never alter a column type in place. Add a new column, backfill, then drop the old one in a separate migration

@migration-template.sql

Glob patterns

Separate multiple patterns with commas.

Pattern	Matches
*.ts	All .ts files in the root
**/*.ts	All .ts files in any directory
src/**	Everything under src/
src/**/*.tsx	All .tsx files anywhere under src/
docs/**/*.md, docs/**/*.mdx	.md and .mdx files under docs/
tailwind.config.*	tailwind.config with any extension

Referencing files from a rule

Write @path/to/file on its own line inside the rule body to include that file's contents in context when the rule loads. This is the recommended way to ship templates and canonical examples: the rule stays short and does not go stale when the referenced code changes.

AGENTS.md

AGENTS.md is a plain Markdown file with no frontmatter. Put it in the project root and Cursor picks it up automatically. It is the right choice when you want readable instructions without per-file scoping, and it is the shared format that other coding agents read too.

# Project Instructions

## Code Style

- Use TypeScript for all new files
- Prefer functional components in React
- Use snake_case for database columns

## Architecture

- Follow the repository pattern
- Keep business logic in service layers

Nested AGENTS.md

Cursor supports AGENTS.md in subdirectories. A nested file applies when Agent works with files in that directory or its children. Instructions are combined with parent files, and the more specific file takes precedence on conflicts.

project/
  AGENTS.md              # Global instructions
  frontend/
    AGENTS.md            # Frontend-specific instructions
    components/
      AGENTS.md          # Component-specific instructions
  backend/
    AGENTS.md            # Backend-specific instructions

CLAUDE.md

Cursor reads CLAUDE.md in the project root the same way it reads AGENTS.md. One difference: CLAUDE.md is always applied to every conversation regardless of any alwaysApply frontmatter. If you need conditional rules, use .cursor/rules/. The CLI also reads both AGENTS.md and CLAUDE.md at the project root and applies them alongside .cursor/rules.

User Rules

User Rules are global preferences that apply in every project. Set them in Customize > Rules (also reachable through Cursor Settings). They are stored on your Cursor account and sync when you sign in on another machine. Use them for communication style and personal conventions, not project facts:

Reply in a concise style. Avoid unnecessary repetition or filler language.
Prefer table-driven tests. Explain reasoning before proposing large changes.

Cursor also loads user rule files from ~/.cursor/rules on the local machine. Unlike account-level User Rules, these do not sync between devices. Neither User Rules nor Team Rules are included in profile exports.

User Rules are used by Agent (Chat) only. They are not applied to Inline Edit (Cmd+K).

Team Rules

Team and Enterprise plans can define rules for the whole organization from the Cursor dashboard. Team Rules are free-form text (no .mdc folder structure) and sync automatically to every member across all repositories.

Option	Effect
Enable this rule immediately	Active on creation; unchecked saves it as a draft
Enforce this rule	Required for all members; cannot be turned off in Customize
Glob pattern (for example **/*.py)	Rule applies only when matching files are in context; without a glob it applies to every conversation

Non-enforced Team Rules are on by default but members can disable them under Team Rules in Customize. Enforced rules are sometimes used for internal compliance workflows; treat them as guidance for the model, not as a security control.

Legacy .cursorrules

A .cursorrules file in the project root still works but is deprecated. To migrate:

1. Create a new rule (command palette, New Cursor Rule)
2. Copy the .cursorrules content into it
3. Set the type to Always Apply (matches the old behavior)
4. Delete .cursorrules

If you want the simplest possible replacement, moving the content into AGENTS.md also works.

Memories

Cursor's current documentation does not describe an automatic, tool-written memory store comparable to Claude Code's auto memory. Everything Agent knows between sessions comes from the rule mechanisms on this page plus what you attach explicitly. Two features cover part of the gap:

• Conversation search as an agent tool: Agent can query your past conversations on its own when it needs context from something discussed earlier.
• @Chats and side chats let you pull a previous conversation into the current one by hand.

If you want Agent to "remember" a decision, write it into a rule or AGENTS.md. You can ask Agent to do that for you: "Add a rule that all API handlers must validate input with Zod."

Loading and precedence

Rules are included at the start of the model context. All applicable rules are merged; when guidance conflicts, earlier sources win:

1. Team Rules (dashboard; enforced or enabled)
2. Project Rules (.cursor/rules/*.mdc, AGENTS.md, CLAUDE.md, .cursorrules)
3. User Rules (account settings and ~/.cursor/rules)

Within project rules, which files load depends on rule type:

• alwaysApply: true rules, CLAUDE.md, and the root AGENTS.md load in every conversation
• globs rules attach when a matching file is in context
• description rules are offered to Agent, which decides whether to fetch them (Agent has a Fetch Rules tool for exactly this)
• Manual rules load only when @-mentioned
• Nested AGENTS.md files load when Agent works in their directory, with the most specific file taking precedence

Check what actually landed in a conversation by clicking the context ring next to the prompt input. The breakdown tray shows a Rules category with the token cost of project and user rules in the current prompt.

Creating rules

Method	How
/create-rule in chat	Type /create-rule in Agent and describe the rule. Agent writes the .mdc file with correct frontmatter into .cursor/rules/
Command palette	Cmd+Shift+P (Ctrl+Shift+P), search New Cursor Rule, name the file, pick the type from the dropdown
Customize sidebar	Customize > Rules > Add Rule. This view also lists every rule and whether it is active
Ask Agent	"Create a rule that..." works in any conversation
Cursor CLI	agent generate-rule (alias agent rule) walks through prompts and writes the file
Remote rule (GitHub)	Add Rule > Remote Rule (Github), paste a repo URL. Cursor scans for .mdc files and syncs them into .cursor/rules/imported/<repoName>/, preserving relative paths

hinweis

Older Cursor versions offered a /Generate Cursor Rules chat command. The current documentation lists /create-rule and the New Cursor Rule palette command instead.

You can also tag @cursor on a GitHub issue or pull request and ask it to update a rule when Agent made a mistake.

What to put in rules

Good rules are focused, actionable, and scoped. Cursor's own guidance: keep each rule under 500 lines, split large rules into composable files, reference files instead of pasting their contents, and write rules the way you would write clear internal docs.

Effective content:

---
alwaysApply: true
---

# Commands

- Tests: npm test
- Lint: npm run lint
- Type check: npm run typecheck

# Conventions

- TypeScript strict mode, never `any`
- All API endpoints validate input with Zod schemas
- Migrations live in src/db/migrations/; never edit an existing one

# Architecture

- Business logic in src/domain/, never import from src/infrastructure/ there
- Use Result<T, E> for error handling, do not throw

What to avoid:

• Copying an entire style guide. Use a linter. Agent already knows common conventions.
• Documenting every command. Agent knows npm, git, pytest, and the like. Document only project-specific ones.
• Edge cases that rarely apply. Keep rules to patterns you hit often.
• Duplicating what the code already shows. Point to a canonical example file with @ instead.
• Vague guidance. "Write clean code" gives the model nothing to act on.

Start small. Add a rule when you notice Agent making the same mistake more than once, and update the rule when it makes a new one.

Constraints, not aspirations

Weak:

Write high-quality, maintainable, well-tested code.

Strong:

Keep functions under 30 lines. Extract helpers for repeated logic.
Never build SQL with string concatenation; use the query builder in src/db/query.ts.
Every new endpoint gets a test in tests/api/ that covers the 400 and 404 paths.

Pick the right type

Situation	Use
Build commands, repo-wide conventions	AGENTS.md or an Always Apply rule
Rules for one part of the codebase (src/api/**, *.tsx)	Apply to Specific Files with globs
Knowledge Agent should reach for only when relevant (a migration playbook, a release checklist)	Apply Intelligently with a clear description
Templates you invoke on demand	Apply Manually, then @rule-name in chat
Personal tone and style	User Rules
Organization-wide standards	Team Rules (enforced if they must not be disabled)

tipp

"Apply Intelligently" rules that are really reusable workflows are often better as Skills. Cursor ships /migrate-to-skills, which converts dynamic rules (alwaysApply: false, no globs) and legacy slash commands into skills.

Monorepo strategy

Use nested AGENTS.md files or globs-scoped rules so package-specific guidance loads only when Agent works in that package:

repo/
├── AGENTS.md                        # Shared: git workflow, CI, formatting
├── .cursor/rules/
│   ├── frontend.mdc                 # globs: packages/frontend/**
│   └── backend.mdc                  # globs: packages/backend/**
├── packages/
│   ├── frontend/
│   │   └── AGENTS.md                # React/TypeScript conventions
│   └── backend/
│       └── AGENTS.md                # Go conventions, API patterns

Both approaches work; nested AGENTS.md is easier to read and is picked up by other agents, while globs rules give you finer file-type control (for example **/*.test.ts across every package). Use .cursorignore to keep irrelevant packages out of context entirely; see Managing Context.

Sharing conventions across tools

If your team uses several AI coding tools, keep one canonical instruction file and point the others at it. AGENTS.md is the natural choice because Cursor, Codex, and others read it natively, and Claude Code can import it:

Tool	Reads natively	Bridge
Cursor	AGENTS.md, CLAUDE.md, .cursor/rules/*.mdc	none needed
Claude Code	CLAUDE.md	CLAUDE.md containing @AGENTS.md, or a symlink
OpenAI Codex	AGENTS.md	none needed
GitHub Copilot	.github/copilot-instructions.md	reference AGENTS.md from it

Put Cursor-only behavior (globs-scoped rules, manual template rules) in .cursor/rules/ and keep AGENTS.md tool-neutral.

Troubleshooting

A rule is not applied. Check the type. An Apply Intelligently rule needs a description. An Apply to Specific Files rule needs a globs pattern that matches a file currently in context. A .md file in .cursor/rules/ is ignored; rename it to .mdc.

Rules seem ignored in Tab or Inline Edit. Expected. Rules apply to Agent only.

A rule from another folder with the same name interferes. Rules are identified by full path and all matching rules apply. Rename or tighten the globs.

Compared with Claude Code

Topic	Cursor	Claude Code
Primary project file	AGENTS.md (plain) or .cursor/rules/*.mdc (frontmatter)	CLAUDE.md, with @file imports
Path-scoped rules	globs: in .mdc frontmatter	paths: in .claude/rules/*.md frontmatter
Model-selected rules	Apply Intelligently via description	No direct equivalent; skills fill this role
Personal, uncommitted project notes	No AGENTS.local.md equivalent; use User Rules	CLAUDE.local.md
Org-wide rules	Team Rules in the dashboard, optionally enforced	Managed policy CLAUDE.md
Auto memory written by the tool	Not documented	~/.claude/projects/<project>/memory/
Precedence on conflict	Team > Project > User	Later, more specific files win (project over user)

See Claude Code Memory for the Claude Code side, and the Cursor overview for a general introduction to the tool.