Generate Cursor rules from your codebase with AI

Scan your code with Gemini to generate .mdc rules automatically.

What rulegen-ai does

Instead of writing rules from scratch, rulegen-ai scans your codebase, analyzes patterns, and uses Gemini to generate structured .mdc rules with frontmatter.

It looks at:

• File structure: Directory layout, naming conventions, imports
• Code patterns: Recurring function signatures, error handling, component structure
• Style choices: Const vs let, arrow functions vs regular functions, naming conventions
• Framework usage: React hooks, Express middleware, TypeScript patterns

The tool generates rules that match what you're already doing, making it easier for Cursor to follow your project's existing conventions.

How to run it

You need a Gemini API key. The free tier works fine for most projects.

export GEMINI_API_KEY=your_key_here
npx rulegen-ai

The command will:

1. Scan your project files (respecting .gitignore)
2. Send code samples to Gemini with a prompt asking for rules
3. Parse the AI response into .mdc files with frontmatter
4. Write the generated rules to .cursor/rules/

What it produces

The tool generates multiple .mdc files, each focused on a specific pattern it found in your code. Example output:

.cursor/rules/
├── typescript-naming.mdc
├── react-components.mdc
├── error-handling.mdc
└── api-routes.mdc

Each file includes:

• Frontmatter: Description and globs that match where the pattern appears
• Rule content: Specific instructions based on what the AI observed in your code
• Examples: Code snippets from your project showing the pattern in use

Customizing the output

The default model is gemini-2.5-flash. You can specify a different model if needed:

npx rulegen-ai --model gemini-2.5-pro

Other options:

• path (positional): Specify a directory to scan (default: current directory). Example: npx rulegen-ai ./my-project
• --format: Output format. Options: cursor (default), claude-md, agents-md, copilot, windsurf
• --max-files: Limit how many files to analyze (default: 50, useful for large codebases)
• --dry-run: Preview rules without writing files
• --verbose: Show which files are sent to Gemini

When to use it vs writing manually

Generated rules are a starting point, not a final product. They work best when:

Review and refine

AI-generated rules need human review. After running rulegen-ai, check the output for:

• Vague instructions: The AI sometimes generates generic rules like "write clean code." Delete or make them specific.
• Incorrect patterns: If your codebase has bad practices, the AI might encode them as rules. Fix these manually.
• Over-scoped globs: Generated globs might be too broad or too narrow. Adjust to match your intent.
• Missing context: Add examples or clarifications where the AI's output is unclear.

Run npx cursor-doctor scan after editing to validate frontmatter and catch quality issues.

Privacy and API usage

The tool sends code snippets to Google's Gemini API. If you're working with proprietary or sensitive code, review your company's policies before using this tool. You can limit what gets scanned with .gitignore or the --path flag to exclude sensitive directories.

Related guides

• How to write Cursor rules that actually work
• Convert rules between Cursor, Windsurf, and Claude Code
• Cursor token budget: how many rules is too many?

npx rulegen-ai