8 steps

Working with your AI without it breaking things

For: Everyone who builds a game by talking to Claude Code, Cursor or another assistant.

An AI is a fast builder with no memory and no eyes. Working with it well means giving it a plan, a way to check its work, and a way back when it breaks something. This path is the habit, not the tools.

The steps, in order

1
Cookbook
How to Vibe Code a Game: One Evening, Start to Finish
Pick one action, have your AI draw the game as a diagram and check it, then build and improve it step by step. Here is a full evening on one game.
Why here: The whole shape of the work first: six gates from idea to a published game, so you know where you are at any moment.
2
Guide
What Is a Core Game Mechanic? A Plain Explanation
A core game mechanic is the action players repeat most, the one the rest of the game supports. How to find yours and describe it to your AI.
Why here: Before the first prompt, name the one thing your game is about. An AI builds whatever you ask; this decides what you ask.
3
Guide
A Simple Workflow for Building a Game With AI
Build in small steps: describe one thing, let your AI build it, have it show what it tried, play the result yourself, and save a working version.
Why here: The daily loop: one feature, a try, a saved version. This is where most people lose days, by skipping the saved version.
4
Guide
My Game Keeps Breaking: Why, and What to Tell Your AI
Your game breaks when you add something because new code touches parts that worked. Change one thing at a time and list what must keep working.
Why here: What to say when a change breaks something else, and why "it should work" from an AI means it did not try.
5
Wiki entry
Boil thousands of repeated error lines down to what you paste to the AI
When something breaks inside the game loop, the same error appears sixty times a second, and the one line that explains the cause scrolls away. Pasting all of it overwhelms the AI. This entry groups lines that differ only in numbers and shows each kind once with a count.
Why here: Errors come in thousands of lines; your AI needs the three that matter. Boiling a log down is a step, not a skill.
6
Story help
Show Me Where That Is Written: Keep Your AI From Guessing
Stop your AI making things up: it quotes the line behind each answer and shows the exact path from a checker. If it cannot, it says it does not know.
Why here: The habit that scales beyond story: ask your AI to show what a claim rests on. Once you expect proof, you get fewer surprises.
7
Wiki entry
Keep the downloaded size of your game under a limit, counted the right way
A web game that is too heavy loses players before the first screen. Adding up file sizes gives the wrong answer, because some files shrink on the way and others do not. This entry shows the AI how to estimate what players really download and name the heaviest files.
Why here: Before publishing, a budget for the download: the one number that decides whether phones even open your game.
8
Guide
What Is MCP? A Plain Guide for Game Makers
MCP is a standard way to plug extra knowledge and tools into your AI. It can give your AI facts and actions, but it does not make its answers right.
Why here: Last, the connection to this wiki: with it your AI looks things up, checks models and saves your project while it builds, instead of guessing.

At the end

You ask for one thing at a time, your AI shows what it checked, and a broken afternoon costs a minute instead of a day.

Say this to your AI

Follow the Spielerando learning path "Working with your AI without it breaking things" (get_path working-with-your-ai) for my game, step by step, and tell me at each step why it comes now.

Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP