anorak999/personal-prompts-by-anorak999

This repository is a collection of three system prompts designed to help developers get more consistent, structured output from AI assistants. The author has written each prompt in five versions, one formatted for each of the major AI models: Claude, ChatGPT, Gemini, Llama, and Mistral. The idea is that each model responds better to its own native syntax, so Claude gets XML, ChatGPT gets Markdown with JSON, and Llama gets special control tokens. The three prompts are named VECNA, BLOB, and ZETA. VECNA handles code performance reviews. You paste in a codebase, and the prompt instructs the AI to find specific performance problems: slow algorithms, memory leaks, blocking database calls, and concurrency bugs. The output comes as a ranked list of findings, each with a before-and-after comparison and implementation guidance. BLOB covers security auditing. It directs the AI to look for injection attacks, authentication bypasses, logic errors, race conditions, and weak error handling. Results include exact file and line numbers, a description of how each issue could be exploited, and a suggested fix. ZETA is a prompt optimizer. If you have a vague request you want to run through an AI, ZETA helps you rewrite it into something more precise and better tailored for the specific model you are using. It applies a four-step process: break down the request, diagnose what is missing, develop a better version, and deliver the result. All three prompts have been compressed by 32 to 62 percent compared to their original versions, according to the README. The author's goal was to keep the same quality of output while spending fewer tokens. Setup is simple: there is nothing to install. The repository is three Markdown files. Each file includes all five LLM variants. You copy the block matching your tool of choice, paste it into your chat session or API call, and follow it with your actual input. The license is MIT.