review-code

Automating a code review against engineering standards involves evaluating diffs across multiple criteria like security, testing, and architecture, then producing a structured report with explicit PASS/WARN/FAIL verdicts. This skill targets staged changes, git ranges, or pull requests to enforce those standards.

Yes, you can review specific git diffs or pull requests for CLAUDE.md adherence. The skill accepts inputs like PR numbers, BASE..HEAD ranges, current branches, or staged changes to evaluate whether the code changes align with your defined engineering conventions.

Automated code review handles intent alignment by optionally incorporating context from plans, architecture decision records, or tickets. This allows the evaluation to verify whether the delivered code changes actually match the intended goals and architectural design before producing a final verdict.

A structured code review report for engineering standards includes identified strengths, per-criterion outcomes, and an explicit verdict with rationale. It evaluates testing, architecture, API contracts, and conventions, while self-filtering low-confidence findings and excluding pre-existing issues.

Yes, the code review evaluation explicitly excludes pre-existing issues found in the codebase. It focuses strictly on the newly introduced code changes referenced by your pull request or git range, ensuring the final verdict reflects only the current diff.

Automated code review for catching security flaws is limited by its self-filtering of low-confidence findings, meaning it only reports high-certainty issues. While it evaluates security as a criterion, it should complement, not replace, dedicated security scanning tools in your workflow.