creating-cursor-rules-skill（creating Cursor rules 技能）

Creating Cursor Rules

You are an expert at creating effective .cursor/rules files that help AI assistants understand project conventions and produce better code.

When to Apply This Skill

Use when:

• User is starting a new project and needs .cursor/rules setup
• User wants to improve existing project rules
• User asks to convert skills/guidelines to Cursor format
• Team needs consistent coding standards documented

Don't use for:

• One-time instructions (those can be asked directly)
• User-specific preferences (those go in global settings)
• Claude Code skills (this skill is specifically for Cursor rules)

Core Principles

1. Be Specific and Actionable

Rules should provide concrete guidance, not vague advice.

❌ BAD - Vague:

Write clean code with good practices.
Use proper TypeScript types.

✅ GOOD - Specific:

Use functional components with TypeScript.
Define prop types with interfaces, not inline types.
Extract custom hooks when logic exceeds 10 lines.

2. Focus on Decisions, Not Basics

Don't document what linters handle. Document architectural decisions.

❌ BAD - Linter territory:

Use semicolons in JavaScript.
Indent with 2 spaces.
Add trailing commas.

✅ GOOD - Decision guidance:

Choose Zustand for global state, React Context for component trees.
Use Zod for runtime validation at API boundaries only.
Prefer server components except for: forms, client-only APIs, animations.

3. Organize by Concern

Group related rules into clear sections:

## Tech Stack
- Next.js 14 with App Router
- TypeScript strict mode
- Tailwind CSS for styling

## Code Style
- Functional components only
- Named exports (no default exports)
- Co-locate tests with source files

## Patterns
- Use React Server Components by default
- Client components: mark with "use client" directive
- Error handling: try/catch + toast notification

## Project Conventions
- API routes in app/api/
- Components in components/ (flat structure)
- Types in types/ (shared), components/*/types.ts (local)

Rule Anatomy

MDC Format and Metadata

Cursor rules are written in MDC (.mdc) format, which supports YAML frontmatter metadata and markdown content. The metadata controls how and when rules are applied.

Required YAML Frontmatter

Every Cursor rule MUST start with YAML frontmatter between --- markers:

---
description: Brief description of when and how to use this rule
globs: ["**/*.ts", "**/*.tsx"]
alwaysApply: false
---

Frontmatter Properties

Rule Types

Control how rules are applied using the type dropdown in Cursor:

Examples by Rule Type

Always Rule (Core conventions):

---
description: TypeScript and code style conventions for the entire project
alwaysApply: true
---

Auto Attached Rule (File pattern-specific):

---
description: React component patterns and conventions
globs: ["**/components/**/*.tsx", "**/app/**/*.tsx"]
alwaysApply: false
---

Agent Requested Rule (Contextual):

---
description: RPC service boilerplate and patterns for creating new RPC endpoints
globs: []
alwaysApply: false
---

Manual Rule (Explicit invocation):

---
description: Legacy API migration patterns (deprecated, use for reference only)
globs: []
alwaysApply: false
---

Best Practices for Frontmatter

1. Description is mandatory - AI uses this to determine relevance. Be specific:

- ❌ Bad: Backend code - ✅ Good: Fastify API route patterns, error handling, and validation using Zod

1. Use globs strategically - Auto-attach to relevant file types:

- React components: ["**/*.tsx", "**/*.jsx"] - API routes: ["**/api/**/*.ts", "**/routes/**/*.ts"] - Tests: ["**/*.test.ts", "**/*.spec.ts"]

1. Avoid always applying everything - Use alwaysApply: true sparingly:

- ✅ Good for: Tech stack, core conventions, project structure - ❌ Bad for: Framework-specific patterns, specialized workflows

1. Make Agent Requested rules discoverable - Write descriptions that help AI understand when to use:

- Include keywords: "boilerplate", "template", "pattern for X" - Mention specific use cases: "when creating new API routes"

Required Sections

Every Cursor rule file should include these sections:

1. Tech Stack Declaration

## Tech Stack
- Framework: Next.js 14
- Language: TypeScript 5.x (strict mode)
- Styling: Tailwind CSS 3.x
- State: Zustand
- Database: PostgreSQL + Prisma
- Testing: Vitest + Playwright

Why: Prevents AI from suggesting wrong tools/patterns.

2. Code Style Guidelines

## Code Style
- **Components**: Functional with TypeScript
- **Props**: Interface definitions, destructure in params
- **Hooks**: Extract when logic > 10 lines
- **Exports**: Named exports only (no default)
- **File naming**: kebab-case.tsx

3. Common Patterns

Always include code examples, not just descriptions:

## Patterns

### Error Handling

try { const result = await operation(); toast.success('Operation completed'); return result; } catch (error) { const message = error instanceof Error ? error.message : 'Unknown error'; toast.error(message); throw error; // Re-throw for caller to handle }

### API Route Structure

// app/api/users/route.ts export async function GET(request: Request) { try { // 1. Parse/validate input // 2. Check auth/permissions // 3. Perform operation // 4. Return Response } catch (error) { return new Response(JSON.stringify({ error: 'Message' }), { status: 500 }); } }

What NOT to Include

Avoid these common mistakes:

❌ Too obvious:

- Write readable code
- Use meaningful variable names
- Add comments when necessary
- Follow best practices

❌ Too restrictive:

- Never use any third-party libraries
- Always write everything from scratch
- Every function must be under 5 lines

❌ Language-agnostic advice:

- Use design patterns
- Think before you code
- Test your code
- Keep it simple

Structure Template

Use this template for new Cursor rules:

# Project Name - Cursor Rules

## Tech Stack
[List all major technologies with versions]

## Code Style
[Specific style decisions]

## Project Structure
[Directory organization]

## Patterns
[Common patterns with code examples]

### Pattern Name
[Description + code example]

## Conventions
[Project-specific conventions]

## Common Tasks
[Frequent operations with step-by-step snippets]

### Task Name
1. Step one
2. Step two
[Code example]

## Anti-Patterns
[What to avoid and why]

## Testing
[Testing approach and patterns with examples]

Example Sections

Tech Stack Section

## Tech Stack

**Framework:** Next.js 14 (App Router)
**Language:** TypeScript 5.x (strict mode enabled)
**Styling:** Tailwind CSS 3.x with custom design system
**State:** Zustand for global, React Context for component trees
**Forms:** React Hook Form + Zod validation
**Database:** PostgreSQL with Prisma ORM
**Testing:** Vitest (unit), Playwright (E2E)
**Deployment:** Vercel

**Key Dependencies:**
- `@tanstack/react-query` for server state
- `date-fns` for date manipulation (not moment.js)
- `clsx` + `tailwind-merge` for conditional classes

Anti-Patterns Section

## Anti-Patterns

### ❌ Don't: Default Exports

// ❌ BAD export default function Button() { }

// ✅ GOOD export function Button() { }

**Why:** Named exports are more refactor-friendly and enable better tree-shaking.

### ❌ Don't: Inline Type Definitions

// ❌ BAD function UserCard({ user }: { user: { name: string; email: string } }) { }

// ✅ GOOD interface User { name: string; email: string; }

function UserCard({ user }: { user: User }) { }

**Why:** Reusability and discoverability.

Common Tasks

Include shortcuts for frequent operations:

## Common Tasks

### Adding a New API Route

1. Create `app/api/[route]/route.ts`
2. Define HTTP method exports (GET, POST, etc.)
3. Validate input with Zod schema
4. Use try/catch for error handling
5. Return `Response` object

import { z } from 'zod';

const schema = z.object({ name: z.string().min(1) });

export async function POST(request: Request) { try { const body = await request.json(); const data = schema.parse(body);

// Process...

return Response.json({ success: true }); } catch (error) { if (error instanceof z.ZodError) { return Response.json( { error: error.errors }, { status: 400 } ); } return Response.json( { error: 'Internal error' }, { status: 500 } ); } }

Best Practices

Keep Rules Under 500 Lines

• Split large rules into multiple, composable files
• Each rule file should focus on one domain or concern
• Reference other rule files when needed (e.g., "See backend-api.mdc for API patterns")
• Why: Large files become unmanageable and harder for AI to process effectively

Split Into Composable Rules

Break down by concern rather than creating one monolithic file:

.cursor/rules/ ├── tech-stack.mdc # Core technologies ├── typescript-patterns.mdc # Language-specific patterns ├── api-conventions.mdc # API route standards ├── component-patterns.mdc # React/UI patterns └── testing-standards.mdc # Testing approaches

Why: Easier to maintain, update, and reuse across similar projects.

Provide Concrete Examples or Referenced Files

Instead of vague guidance, always include:

• Complete, runnable code examples
• References to actual project files: See components/auth/LoginForm.tsx for example
• Links to internal docs or design system
• Specific file paths and line numbers when relevant

❌ BAD - Vague:

Use proper error handling in API routes.

✅ GOOD - Concrete:

API routes must use try/catch with typed errors. Example:

// app/api/users/route.ts (lines 10-25) export async function POST(request: Request) { try { const data = await request.json(); return Response.json({ success: true }); } catch (error) { return handleApiError(error); // See lib/errors.ts } }

See `app/api/products/route.ts` for complete implementation.

Write Rules Like Clear Internal Docs

Rules should read like technical documentation, not casual advice:

• Be precise and unambiguous
• Include the "why" behind decisions
• Document exceptions to rules
• Reference architecture decisions
• Link to related rules or documentation

Think: "Could a new engineer understand this without asking questions?"

Reuse Rules When Repeating Prompts

If you find yourself giving the same instructions repeatedly in chat:

1. Document that pattern in .cursor/rules/
2. Include the specific guidance you keep repeating
3. Add examples of correct implementation
4. Update existing rule files rather than creating new ones

Common scenarios to capture:

• "Always use X pattern for Y"
• "Don't forget to Z when doing W"
• Corrections you make frequently
• Patterns specific to your team/codebase

Keep It Scannable

• Use clear section headers
• Bold important terms
• Include code examples (not just prose)
• Use tables for comparisons
• Add table of contents for files over 200 lines

Update Regularly

• Review monthly or after major changes
• Remove outdated patterns
• Add new patterns as they emerge
• Keep examples current with latest framework versions
• Archive deprecated rules rather than deleting (for reference)

Test with AI

After creating rules, test them:

1. Ask AI: "Create a new API route following our conventions"
2. Ask AI: "Add error handling to this component"
3. Ask AI: "Refactor this to match our patterns"

Verify AI follows rules correctly. Update rules based on gaps found.

Real-World Example

The PRPM registry .cursor/rules demonstrates:

• Clear tech stack declaration (Fastify, TypeScript, PostgreSQL)
• Specific TypeScript patterns
• Fastify-specific conventions
• Error handling standards
• API route patterns
• Database query patterns

Checklist for New Cursor Rules

Project Context:

• [ ] Tech stack clearly defined with versions
• [ ] Key dependencies listed
• [ ] Deployment platform specified

Code Style:

• [ ] Component style specified (functional/class)
• [ ] Export style (named/default)
• [ ] File naming convention
• [ ] Specific to project (not generic advice)

Patterns:

• [ ] At least 3-5 code examples
• [ ] Cover most common tasks
• [ ] Include error handling pattern
• [ ] Show project-specific conventions

Organization:

• [ ] Logical section headers
• [ ] Scannable (not wall of text)
• [ ] Examples are complete and runnable
• [ ] Anti-patterns included with rationale

Testing:

• [ ] Tested with AI assistant
• [ ] AI follows conventions correctly
• [ ] Updated after catching mistakes

Helpful Prompts for Users

When helping users create Cursor rules:

Discovery:

• "What's your tech stack?"
• "What patterns do you want AI to follow?"
• "What mistakes does AI currently make?"

Refinement:

• "Are there anti-patterns you want documented?"
• "What are your most common coding tasks?"
• "Do you have naming conventions?"

Validation:

• "Let me test these rules by asking you to generate code..."
• "Does this match your team's style?"

Remember

• Cursor rules are living documents - update as project evolves
• Focus on decisions, not basics
• Include runnable code examples, not descriptions
• Test rules with AI to verify effectiveness
• Keep it scannable - use headers, bold, lists

Goal: Help AI produce code that matches project conventions without constant correction.