Claude Code Mods

After Claude changes files or runs commands, an Explain it button appears. Press it and a side panel walks through what just happened: a one-line summary, the steps in plain English, the same steps the way an engineer would say them, why it matters, and whether I need to do anything. It ends with two to four real technical words to learn, and remembers the ones I already know so it teaches new ones each time. It's how I'm learning to read my own projects. Everything stays on my computer.

Before a project goes live, I say "ship check" and Claude runs a pre-launch checklist: does it work, does it work on a phone, are any secret keys exposed, are the core security rules on, and is it polished. It fixes the small things and hands back a plain-English report card. A built-in guard won't let Claude publish until the check passes. Fun fact: the guard blocked its own first upload to GitHub until Ship Check had run on it.

Codex and Claude work on the same project by passing a written hand-off note, like a relay baton. If one runs out of usage, the other picks up where it stopped. Or one does the work and the other checks it. I tested whether the checker just agrees: when it read the other one's report first, it sometimes approved unfinished work. When it looked at the work on its own first, it kept its own view every time, so that's the default. It asks me when a decision is mine to make, and it runs on the plans I already pay for, with no API keys.