Code Review & Quality
Reviews every change before merge across five axes — correctness, readability, architecture, security, and performance — approving when a change definitely improves overall code health rather than demanding perfection. Its distinctive machinery is severity-labelled findings (Critical / no-prefix Required / Nit / Optional / FYI) so authors can tell what is mandatory from what is discretionary, paired with change-sizing norms (~100 lines is reviewable, ~1000 must be split), splitting strategies (stack, file-group, horizontal, vertical), and review-speed expectations (respond within one business day).
It emits an artifact-review-report and can delegate the actual reviewing to the addy-code-reviewer persona. The skill applies pattern-anti-rationalization — its rationalization tables and structural-remedy prompts exist to stop reviewers rubber-stamping or softening real issues.
See Also
- addy-code-reviewer — the persona this delegates the review to.
- bmad-code-review — BMAD’s counterpart; adversarial parallel review layers over one diff (both emit an artifact-review-report).
- artifact-review-report — the labelled findings it produces.
- stage-review — the canonical stage this implements.