Skip to content

AI App Starter: six fixes from running it on a live site

Latest

Choose a tag to compare

@TechNomadCode TechNomadCode released this 17 Oct 07:36

I've run the AI App Starter on promptquick.ai for three weeks. These are the problems that actually broke the workflow, now fixed in the starter.

AI App Starter

  • Emails from staging and production no longer block each other when both use one email account.
  • The background-job scheduler cleans up its own run history, so it can't creep toward a free database plan's storage limit.
  • A new .gitattributes keeps line endings the same on every OS, so the automatic checks pass on Windows too.
  • Risky pieces of work keep their requirements until their independent review is done, so the reviewer can check against them.
  • The setup steps the agent writes for you are click by click, in a working order, and say exactly where each setting goes and when a redeploy is needed.
  • The main instructions file gets a size limit it can actually meet.

READMEs

  • They now say only what the starter does: instructions and permission rules, no code. Codex is marked untested.
  • The hosted staging and production pipeline is now marked as tested, on promptquick.ai.
  • The independent review is described fully: sign-in, payments, database changes and live data.

Projects you already set up keep the version they started from. To use these fixes there, ask your agent to compare its .kit/ folder with this release.