Architecting Code with LLMs: Advanced Prompts for Refactoring, Unit Testing & Security Audits

Software engineers who use AI simply as an auto-complete tool are missing 90% of its utility. When properly prompted with contextual boundaries and architectural directives, frontier language models excel at complex static code analysis, identifying subtle concurrency deadlocks, refactoring monolithic classes into SOLID architectures, and writing robust unit test suites covering tricky boundary conditions.

1. The Principal Engineer Code Audit Prompt

To prevent the AI from generating lazy mock implementations (such as // TODO: implement logic here), your prompt must enforce strict production-grade expectations.

You are a Principal Software Architect conducting a rigorous peer code review.
I will provide a code module in <source_code>.

Audit Requirements:
1. Architectural Review: Identify violations of Single Responsibility, Open/Closed, and Dependency Inversion.
2. Performance & Concurrency: Flag memory leaks, unclosed connections, N+1 database queries, and race conditions.
3. Security Audit: Scan for SQL injection, unsanitized deserialization, prototype pollution, and timing attacks.
4. Production Refactor: Provide the complete, refactored code without omitting any lines or using placeholder comments.

Negative Directives:
- Do not use '// ... rest of code stays the same'. Output the entire runnable module.
- Adhere strictly to idiomatic typing and modern syntax standards for the language.

<source_code>
[PASTE YOUR CODE HERE]
</source_code>

2. Automated Unit Test Generation: The Mutation Strategy

Most AI-generated unit tests test only the 'happy path'—verifying that valid inputs return expected outputs. Professional engineering requires stress-testing malicious inputs, null pointers, timeout exceptions, and boundary limits.