arnie936/llm-prompting-tests

This repository is a collection of 16 challenging prompts designed to test how well AI models, coding assistants, and AI agents perform on complex technical tasks. The prompts are written primarily in German with some in English, and each one asks an AI to build something substantial and self-contained from scratch. The prompts cover a wide range of simulation, game development, and creative engineering tasks. Examples include building a working beehive simulation as a single HTML file, complete with hexagonal cells, bee roles, resource management, and environmental factors. Another asks for a full 3D motorcycle racing game with AI opponents, collision physics, and a minimap, also in a single HTML file. A first-person shooter prompt specifies mouse-look controls, enemy AI with patrol and chase behavior, multiple weapons, and a self-diagnosis check at startup. One prompt asks for a skateboarding game written entirely in a single C++ source file. Another requests a tower defense game with a built-in map editor and a post-wave balancing indicator. A recurring theme across the prompts is the requirement to pack everything into a single file with no external dependencies. Several prompts also ask the AI to include a short section in the code explaining design decisions, and some ask for a self-diagnostic check at startup to confirm that core systems initialized correctly. These constraints test whether a model can produce code that is not just functional but also organized and self-aware about its own outputs. The collection appears to be a personal benchmark suite rather than a library or tool. There is no code in the repository itself, only the prompt texts. Each prompt is formatted as a standalone specification that could be pasted directly into any AI chat interface. The full README is longer than what was shown.