AI Agents Built a Browser MMO in One Day: Auditing All 115 Commits
Claude agents made 115 commits on a browser MMO in one day, and 38 were game code. What ran at midnight, what broke in the next three days, how to audit yours.
9 posts
Claude agents made 115 commits on a browser MMO in one day, and 38 were game code. What ran at midnight, what broke in the next three days, how to audit yours.
My CLAUDE.md grew from 10.9 KB to 46.5 KB in 38 days, one agent mistake at a time. Seven incidents, the rules they left, and how to word a rule that holds.
A daemon that read 16 GB in 8.5 hours, a spinner that leaked 19 GB, a Bun handle leak and 2 million inotify watches: how each was measured, and the fix.
I ran the same 3D asset briefs on Claude Opus, Sonnet and Haiku at every reasoning effort level. xhigh cost more, took 2.3x as long and looked no better.
How I answer Claude Code's permission prompts and questions from an iPhone: one hub for every session, Postgres search over 9.4 GB of transcripts, QR pairing.
An audit of 1,086 Claude Code sessions: 72% of the spend was cache reads and 53% paid for turns that only read code. The checks to run on your own transcripts.
expo-wsl-ios builds your Expo app and installs it on a real iPhone, from Windows, with no Mac and no EAS. It compiles in WSL against Apple's real iOS SDK, signs with your App Store Connect key, and installs over USB.
I've been building alt.tv.rewind: take the Internet Archive's dumps of the alt.tv. newsgroups, line every thread up against the original air dates, and let you click an episode of Seinfeld and read what Usenet said the…
Most of my development runs through Claude Code, and a lot of it ends in a browser: does the component render, does the flow click through, is the bug fixed. For months the answer was the Claude in Chrome extension — an…