Diagnose before fixing (break the AI fix loop)
Forces evidence before edits. Use it the second time the same error shows up.
For people building with Cursor, Claude Code, Codex, Lovable, Bolt, Replit & v0
Plain-English explanations for the errors AI-built apps keep hitting, each with a fix prompt you paste straight back into your AI tool. Plus 28 starter prompts for the next build.
Try: · · ·
.env files stay on your machine. Your host only knows the variables you add in its dashboard.
Row Level Security is on for that table, but no policy allows this user to insert (or update) this row.
The browser refused to give your page the response because the API didn't say your site is allowed to read it.
The first render on the server and the first render in the browser produced different HTML.
Anything in client-side code, including NEXT_PUBLIC_ and VITE_ variables, is downloadable by anyone. Treat that key as compromised.
Each blind retry adds more code and more guesses. Stop patching and get a diagnosis first.
Forces evidence before edits. Use it the second time the same error shows up.
A checklist-driven review of the mistakes that most often leak data in AI-built apps.
Sign-up, login, and a dashboard where each user only sees their own records, with RLS done right from day one.
A minimal Model Context Protocol server that exposes one useful tool to Cursor, Claude, and other MCP clients.
A complete small game loop: input, movement, collisions, score, and restart. No engine needed.
Browse NASA's APOD by date, with explanation text, favourites, and a random button.