12 Prompts for Claude Code: From Refactoring to Architecture

Introduction

Claude Code is an AI-powered coding assistant developed by Anthropic, designed to integrate seamlessly into your development workflow. Unlike generic chatbots, Claude Code operates directly in your terminal, understands your codebase contextually, and can perform complex tasks like refactoring, code review, and architectural design. This article presents 12 carefully crafted prompts organized into three categories: Basic, Advanced, and Expert. Each prompt includes a clear task, the exact prompt text, and a real-world example result. Whether you are a junior developer looking to clean up your code or a senior engineer designing system architecture, these prompts will help you get the most out of Claude Code.

Category 1: Basic Prompts

1. Code Review

• Task: Review a specific function or file for bugs, style issues, and improvements.
• Prompt: "Review the function processPayment in payment.js for potential edge cases, error handling, and adherence to JavaScript best practices. List any bugs or improvements."
• Example Result: Claude Code scans the file and returns a structured report: "Issue 1: Missing check for null amount — could cause runtime error. Suggestion: Add early return if typeof amount !== 'number'. Issue 2: Variable tax is not declared with let or const — implicit global. Fix: Use const tax = ...."

2. Refactor Legacy Code

• Task: Convert a long function into smaller, reusable functions.
• Prompt: "Refactor the function handleUserSignup in auth.js into smaller helper functions. Keep the original behavior, but make the code more modular and readable."
• Example Result: The original 80-line function is split into validateEmail, hashPassword, createUserRecord, and sendWelcomeEmail. Claude Code provides the refactored code and a brief explanation of each new function.

3. Add Unit Tests

• Task: Generate unit tests for an existing module.
• Prompt: "Write comprehensive unit tests for the module utils/stringHelpers.js using Jest. Cover normal cases, edge cases, and error scenarios."
• Example Result: Claude Code outputs a test file with 15 test cases, including tests for empty strings, special characters, and type coercion errors, along with comments explaining each test.

Category 2: Advanced Prompts

4. Design a Module from Scratch

• Task: Create a new module based on high-level requirements.
• Prompt: "Design a caching module in Python that supports TTL, LRU eviction, and thread-safe operations. Provide the class structure, key methods, and a usage example."
• Example Result: Claude Code returns a complete Cache class with methods get, set, delete, and clear, using collections.OrderedDict for LRU and threading.Lock for safety. It also shows how to instantiate it: cache = Cache(maxsize=100, ttl=300).

5. Refactor with Design Patterns

• Task: Apply a design pattern to improve code structure.
• Prompt: "The current code in notifications.py uses multiple if-elif chains to handle different notification types (email, SMS, push). Refactor it using the Strategy pattern."
• Example Result: Claude Code creates a NotificationStrategy interface, concrete classes (EmailStrategy, SMSStrategy, PushStrategy), and a NotificationContext class. The original 50-line conditional chain becomes a clean, extensible 10-line dispatching mechanism.

6. Cross-File Refactoring

• Task: Rename a function across multiple files.
• Prompt: "Rename the function fetchUserData to getUserProfile in all files under the src/ directory. Update all imports and usages."
• Example Result: Claude Code identifies all 12 files referencing fetchUserData, performs the rename, updates imports, and shows a diff of changes. It also checks for any potential conflicts.

7. Performance Optimization

• Task: Identify and fix performance bottlenecks.
• Prompt: "Analyze the dataProcessor.py file for performance issues. The function processLargeDataset takes ~10 seconds for 100k records. Suggest and implement optimizations."
• Example Result: Claude Code identifies that the function uses nested loops with O(n²) complexity and suggests using a hash map for lookup. It rewrites the function to achieve O(n) complexity and estimates a 5x speed improvement.

Category 3: Expert Prompts

8. Architectural Design Proposal

• Task: Design a microservices architecture for a given system.
• Prompt: "Propose a microservices architecture for an e-commerce platform with services for product catalog, user management, orders, and payments. Include service boundaries, communication protocols (REST vs. message queue), and data storage strategy."
• Example Result: Claude Code produces a detailed architecture document with a diagram description (ASCII), tables for each service, and recommendations for using PostgreSQL for transactional data and Redis for caching. It also explains when to use synchronous vs. asynchronous communication.

9. Database Schema Migration

• Task: Plan a migration from a monolithic database to a sharded setup.
• Prompt: "Design a migration plan for a PostgreSQL database that currently stores all user data in a single table. The database is experiencing performance issues due to 10 million users. Propose a sharding strategy (e.g., by user_id hash) and the steps to migrate without downtime."
• Example Result: Claude Code outlines a 5-phase plan: (1) create shard databases, (2) implement a proxy layer for routing, (3) backfill data in batches, (4) dual-write for consistency, (5) cut over. It also provides sample SQL for shard key logic.

10. Security Audit

• Task: Perform a security review of an authentication module.
• Prompt: "Audit the authentication module in auth_service.py for common vulnerabilities: SQL injection, XSS, CSRF, and insecure password storage. Provide a report with severity levels and fixes."
• Example Result: Claude Code returns a report with three findings: (1) Critical — raw SQL concatenation allows injection, fix with parameterized queries. (2) High — passwords stored as MD5, replace with bcrypt. (3) Medium — missing CSRF token on login endpoint. Each finding includes a code snippet for the fix.

11. API Contract Design

• Task: Design a RESTful API with OpenAPI specification.
• Prompt: "Design a RESTful API for a task management system. Include endpoints for CRUD on tasks, user assignment, and status transitions. Provide the OpenAPI 3.0 spec in YAML format."
• Example Result: Claude Code generates a complete OpenAPI spec with 10 endpoints, request/response schemas, authentication (Bearer JWT), and examples. It also explains design decisions like using PATCH for partial updates and HTTP status codes.

12. Full System Refactoring Plan

• Task: Create a step-by-step plan for refactoring a monolith to microservices.
• Prompt: "We have a monolithic Ruby on Rails application with 500k lines of code. Create a refactoring plan to extract the billing module into a separate microservice. Include risk assessment, testing strategy, and rollback plan."
• Example Result: Claude Code produces a comprehensive plan: (1) Identify bounded context using Domain-Driven Design. (2) Extract billing code into a new repository. (3) Implement a facade pattern in monolith to route calls. (4) Set up CI/CD for the new service. (5) Run in parallel for two weeks with feature flags. (6) Rollback plan: revert to facade and disable new service. It also includes a risk matrix table.

Conclusion

Claude Code is more than a code generator — it is a collaborative partner that can help you review, refactor, and design software at every level. The 12 prompts in this guide range from simple code reviews to full architectural planning, giving you a toolkit for any coding challenge. Start with the basic prompts to build confidence, then move to advanced and expert prompts as you become more familiar with Claude Code's capabilities. Remember: the quality of the output depends heavily on the clarity of your prompt. Be specific, include context, and always review the generated code before deploying. Try these prompts in your next project and see the difference firsthand.

This article is based on official documentation from Anthropic (docs.anthropic.com) and community best practices as of July 2026.