System request for ChatGPT and Claude: debug, review, refactoring, and security code in one.

Systematic prompt for ChatGPT and Claude: debug, review, refactoring and code security in one

If you've ever asked ChatGPT to fix a bug and it rewrote the entire file and broke three things instead — you'll understand why this prompt exists.

What it is

КОДЕМИУРГ GPT X is a system prompt that turns any LLM into a disciplined senior developer. Not a chatty assistant that explains the obvious, but an engineer who reads the task, understands the context, and does exactly what was asked — no more, no less.

What it can do

The prompt works in 16+ modes that switch automatically based on how you phrase your request:

• DEBUG / PATCH — finds the root cause and delivers a minimal fix without touching the rest of the code
  
• CODE REVIEW — checks logic, edge cases, security, and architectural risks using a structured checklist
  
• SECURITY — detects SQL injections, XSS, CSRF, path traversal, data leaks, and weak authorization
  
• REFACTOR / OPTIMIZATION — improves readability and performance while preserving behavior
  
• TESTING — writes tests for your specific framework with coverage for normal cases and edge cases
  
• ARCHITECTURE / DATABASE / API — designs structure with real risks in mind
  
• FRONTEND / BACKEND / DEVOPS — separate modes for each layer of the stack
  
• DOCUMENTATION — generates README with commands, examples, and common errors
  

How it's different from a regular "write me some code" prompt

Most prompts just ask the model to be "helpful" — and get verbose answers packed with theory. This one works differently: strict rules against hallucinations (no inventing API methods, library versions, or nonexistent parameters), against rewriting more than needed (a patch is a patch, not a full project refactor), and against breaking business logic.

A dedicated INPUT SAFETY block protects against prompt injection — if you paste someone else's code with malicious comments, the model won't execute instructions hidden inside it.

Who it's for

— PHP/Python/JS developers who want fast, accurate answers without filler
— Anyone doing AI-assisted code review
— Solo developers with no team who need a "second pair of eyes" on security
— Everyone tired of responses like "here's a full rewrite using a new framework"

How to use it

Paste the prompt as a system instruction in ChatGPT, Claude, or any LLM that supports system prompts. Then just write your task in your own words — the mode is detected automatically. Want only a diff — say "diff only". Want a detailed breakdown — say "detailed". The prompt listens.

Battle-tested on real tasks: PHP backend debugging, Redis security system review, CLI utility test writing.

Download the Prompt.