Lite is free. Pro adds the quality layer on top.
Review beforeyou deploy.
Ten minutes of review before a deploy catches most of what an AI quietly got wrong.
// the steps
Six steps
01 Read the whole diffRun git diff against the last good version. Look for files you did not expect, deleted code and anything that reads like a placeholder.
02 Run the tests and the buildRun them yourself and read the end of the output. A green line you did not see is not a result.
03 Open the pages at three widthsCheck phone, tablet and desktop. Scroll the full page and press every button.
04 Search for secretsLook for keys, tokens and passwords in the diff and in tracked env files. Check nothing sensitive is in a public folder.
05 Try the failure pathSubmit an empty form, a very long value and a wrong password. Unhappy paths are where AI code is thinnest.
06 Know how to go backBefore you deploy, note the previous good version and how to roll back to it. Do not deploy on a Friday evening.
// with adams
Where Adams fits
A reviewer that did not write itProPro hands the change to a separate reviewer agent, ranks findings by severity and fixes only on your yes. The page check covers the three widths.
// faq
Questions
Is this enough for a serious product?
It is a floor. Add automated tests, monitoring and, for sensitive data, a professional security review.
How long should this take?
Ten to thirty minutes for a normal change. More for anything touching money or personal data.
Who should review?
Someone other than the author. When that is impossible, a fresh AI session with no memory of the work is better than none.
// related