How a Severity Rubric Stops AI Code Reviewers From Crying Wolf

A DEV Community article argues that AI code reviewers become ignored because they flag too many low-priority issues, and proposes a structured fix using a three-tier severity rubric. The rubric defines blockers as bugs requiring a named concrete input — such as data loss or a security hole — while major findings cover real-world edge cases like unhandled errors or race conditions. Minor findings are limited to five specific, incident-prone patterns including missing timeouts and swallowed exceptions, deliberately excluding style or formatting concerns. A key rule requires that any finding lacking a concrete trigger — a specific, realistic input — must be deleted rather than reported. If no finding meets the bar, returning an empty review is treated as the correct and expected output for most code diffs.