# Review checklist

The agent will say it's done. Everything committed, everything pushed. Then you actually look, and the text you asked to make larger has got smaller. That's normal. The agent does the work, and the checking is still yours. Budget real time for it: on a build day, the review can take as long as the building.

## Every task

- [ ] **Read the summary**, then check it matches what I asked for, not a nearby thing.
- [ ] **Look at which files changed.** Any surprises? Ask why before accepting.
- [ ] **Run it myself.** Don't just take its word that the tests pass.
- [ ] **Check "done when"** from the brief, exactly as written.
- [ ] **Check the next-door stuff.** Changes break neighbours: the page beside it, the button below it.
- [ ] **Phone and laptop,** light and dark mode if the app has them.
- [ ] **The empty state:** what a brand-new user sees.

## Before anything goes live

- [ ] No API keys, passwords, or personal data anywhere in the code (ask: "Search the project for anything that looks like a secret").
- [ ] New packages: I know what each one is for.
- [ ] Nothing I asked it not to change has changed.
- [ ] The preview link works when I'm logged out, in a private window.

## Asking it to check itself

It helps, but it doesn't replace you:

```
Review the changes in this branch as if you were a careful senior
developer seeing them for the first time. List anything that's broken,
risky, confusing, or not what the task asked for. Don't fix anything
yet; just list it, most important first.
```

## Review log

Task:

What I found:

Sent back with:

Rule I added to the memory file so it doesn't happen again:
