Crafting Production-Grade Prompts

To get consistent, high-quality results from Large Language Models (LLMs), prompts must provide Role, Context, Task, Constraints, and Output Format.

The 5-Part Prompt Framework

1. Persona / Role: Define who the AI is acting as.
2. Context: Give relevant background information.
3. Instruction / Task: Clearly define the specific task to perform.
4. Constraints: State what NOT to do and limits (e.g. no markdown, max 100 words).
5. Output Specification: Define the exact structure (JSON, Markdown table, YAML).

Example: Weak Prompt vs Engineered Prompt

Weak Prompt

Write a login function for PHP.

Production Prompt Template

You are a senior cybersecurity and backend PHP 8.4 architect.

Task: Write a secure user authentication function that verifies a user's password using password_verify() and issues a secure session.

Constraints:
- Use PHP 8.4 strict typing (declare(strict_types=1);).
- Implement rate limiting prevention against brute force attacks.
- Do not expose detailed error messages to the client.
- Return response as JSON containing: { "success": bool, "message": string }.