On Friday evening I gave myself one evening to build a Halloween game from scratch. About eight hours later, Elijah's Halloween Quest was live.
You can play it at halloweenquest.app. It is a run-and-gun platformer that works in a browser on a phone or a computer. You collect sweets, change costumes, dodge the monsters and try to get home before midnight.
I want to be clear about the point. It was a bit of fun. I wanted to see what one evening could produce now.
What is in it
The game has multiple levels, which it calls streets. They run from a suburban cul-de-sac through a churchyard, a farm and a funfair to a mansion and a town square past midnight. There are eight costumes, each with its own move. A secret Ghost Train appears only once you have found six hidden extras on the first street. Everything has music, sound effects and a whispering narrator, and all of the pictures are custom.
Three tools made it.
- Claude Code wrote the game.
- ElevenLabs made the sound. It composed a 60-second instrumental for each street, generated the sound effects and spoke 63 narrator lines.
- Higgsfield made the pictures. The characters and props are 3D-rendered, the title screen is a poster, the backgrounds scroll in three layers and the characters have animated run cycles.
I did not draw, compose or record any of it myself.
How the evening went
The project starts with a design document, a style guide and an audio brief. Each tool worked from the same brief. I think that is why three different systems produced something that looks and sounds like one game.
The log shows 33 commits on Friday. The first was at 17:08 and the last at 23:47. By midnight it was a working game with music, a narrator and a secret level.
A good share of those commits are repairs. A hit flash that never faded. A hairline seam in the sky. A werewolf that could fall out of the world. Faster loading, and touch buttons that had to be laid out before the game started. The first version was not the one you can play now. A lot of the evening went on finding what was wrong and fixing it.
This morning I added four more streets: the toy shop, the frozen pond, the haunted school and the rooftops.
Does it hold up?
It is a finished game, not a prototype. Five play testers tried it and all five liked it. That is a small group, so I will not claim more than that. Chrome and Edge are the best tested browsers for controllers.
What changed
Games are not easy now. This one still took 34 commits of fixing. What has changed is the price of finding out whether an idea is any good. Last Friday the answer cost an evening.
If finding out costs an evening now, which of your ideas is still waiting for a month it does not need?