An Unscoped AI Reviewer Will Invent Problems

Ask an unscoped reviewer for gaps and it finds some, whether or not any exist. That's what you asked it to do. A common shape this takes: a review comes back with a dozen findings, you fix all of them dutifully, and the simple feature you started with now has extra abstraction layers and tests for cases that cannot occur.

Nothing malfunctioned there. The reviewer did exactly what an unscoped "find gaps" prompt asks for.

Scope it: "flag only gaps that affect correctness or a stated requirement." Then treat the output as a list to triage, not a list to complete.

In multi-agent setups this compounds: several unscoped reviewers each inventing a little produces a lot of defensive code nobody needed.