A second opinion

Think of it when a change is ready to merge, especially one the author is sure about.

In everyday life

Like a proofreader who did not write the article. The author reads the sentence they meant to write; a fresh reader sees the word that is actually missing.

One case

On Monday evening, a developer ships a change to a 120-room hotel’s booking site: a room is held for 15 minutes while the guest pays. Nobody reads the diffThe lines a change adds and removes, shown side by side. Reviewers read the diff to see what changed., and nothing releases the hold when a guest closes the tab. By Friday, 40 empty rooms show as taken.

Each abandoned payment held a room for good

The case, cell by cell1 cell = 1 day

Rooms really booked

7075727880

Rooms shown as taken

708493108120

Empty rooms shown as taken

People who read the change

One reviewer question, “what releases the hold?”, would have caught it on Monday.

Every case

The author is the worst person to check a change, because they read what they meant to write. A reviewer comes without that picture, so they see what is actually there: a missing case, a change bigger than the task, a rule the team agreed on. Review works best with two questions: does it do what was asked, and does it follow the team’s rules? Give the reviewer the change and the task, not the author’s reasoning, or the reviewer inherits the author’s blind spots.

The word

Engineers call this

code reviewA second person or agent reads a change before it joins the main code, and checks it against what was asked and the team’s rules.

Have every change read by someone who did not write it, before it merges. Ask: does it do what was asked, and nothing more?

Where this idea leads

happens on aA change asking to joinFoundations, How teams ship
is done for an agent byAn agent that did not write itSteering, Check it
does not replaceProve it worksHabits, While you build
leaves the repeatable checks toSame checks, every time

Skills that encode it

code-review Matt PocockReviews the diff on two separate axes, in separate sub-agents: does it follow the repo’s standards, and does it match the spec?
interrogate pstackAdversaries try to break the change, and a finding counts only when a run against the real code proves it.
council tstackThree models from three providers review the work, and the author keeps the reviews from growing the scope.