Code from the posts

Each of these is generated from the file that actually runs — not pasted from it — so a snippet here cannot drift from the thing it was taken out of. Every one is written up in full on indiecore.net.

• Static site → Cloudflare Workers with gated GitHub Actions deploys

  A build-verify-deploy pipeline on Cloudflare's free tier, and the five traps that cost me an evening — including one that silently takes your staging URL offline.

  ci-cd.yml wrangler.jsonc

  Read the code Deploying a static site to Cloudflare Workers

• Announce a blog post on LinkedIn from CI — the API script and the workflow

  The LinkedIn API call that publishes a post is six lines. The token that expires every 60 days with no refresh, the read scope you cannot have, and a filename that became a dead link took the rest of the afternoon.

  linkedin.mjs linkedin.yml

  Four things break when CI posts to LinkedIn

• Build gates for a static site — output verification and a Lighthouse budget

  Reading every diff at generation speed isn't review, it's skimming. Here is the 169-line verifier that gates this site, and the bug behind every check in it.

  lighthouse-check.mjs verify.mjs

  Read the code I don't review AI code. My build does. The check that passed while broken Cloudflare injects a beacon. My CSP said no.

• Responsive image pipeline — Chrome's byte budget, measured sizes, one cache

  Improve image delivery" makes two complaints in one panel. Both have exact arithmetic behind them, readable straight out of the JSON report.

  image-pipeline.mjs responsive-markup.mjs

  Read the code Lighthouse's image budget is w × h ÷ 6

• Accessible image lightbox — focus trap, restore, and the visibility trap

  My lightbox was invisible and still in the tab order. Fixing that broke focus handling silently, because focus() on a hidden element reports nothing.

  lightbox.css lightbox.js

  Read the code opacity 0 does not hide anything

• Rating puzzle difficulty by the decisions a level forces

  Grid size is the obvious difficulty proxy and it is wrong. Measuring the branching factor of the solve gives a rating that survives contact with players.

  summarise.js tiers.js walk.js

  Read the code Rating puzzle difficulty by decisions

• Campaign ordering — difficulty rates a level, progression places it

  Sorting a level library easiest-to-hardest gives a campaign that is flat for a hundred stages and then a wall. What replaced the sort.

  beats.js envelope.js tones.js

  Read the code Difficulty rates a stage, progression places it

• A Vite plugin that shrinks a JSON content pack into the build output

  392 levels of JSON, pretty-printed for review and minified for the store. The Vite plugin that does it, and the hook choice that hid a real error.

  shrink-stage-pack.js

  Read the code Shipping a 4 MB level pack

• Cutting a Capacitor Android download in half — R8, ProGuard and the real delivered size

  13.1 MB down to 5.9 MB delivered. R8 is off in the project Capacitor generates, and one library's consumer ProGuard rule was pinning 9,304 classes.

  delivered-size.mjs drop-web-firebase.js proguard-rules.pro release-build.gradle

  Read the code Cutting a Capacitor Android build in half

• Conformance suite for generated game content — walk every level before shipping it

  Unit-testing the game does not tell you whether the 392 levels in this build are solvable. So the suite plays every one of them.

  pack-suite.js

  Read the code Testing a level pack you did not write

• A mobile game coin economy priced off its own difficulty model

  Levels differ 13x in the thinking they demand, so a flat coin reward is wrong on nearly all of them. Tying payouts to the difficulty walk.

  economy.js

  Read the code Pricing an economy off its difficulty model