VibeCity

The agent was consuming tokens and contexts faster then you think, planning new features took ages. Even the result was always good i came to the conclusion that my approach does not scale very well. Today morning, i “prompted” myself a set of rules and procedures that the agents have to follow. Now the agents learn about patterns/apis/libs and store the condensed result inside the project, i can interactively plan features ahead and the result is a set of simpler tasks with dependecies on other tasks and requirements. Those also get stored inside the project. then of course i can yolo mode trough the tasks to implement actual code. the agents use git as they should, and i have implemented even a simple locking mechanism. I have planned out the terrain system and an editor. the whole process took 2 hours and now claude and gemini have been programming for over 4 hours. They are still not finnished. I hope i can show a screenshot later