Code Review

Input: a fixed Git comparison point or clearly bounded working diff, applicable specs/ticket, repository standards, and verification output. Output: actionable findings for the caller.

Pin the diff before review. Check two separate axes: Spec (missing, incorrect, or extra behavior) and Standards (correctness, maintainability, conventions, security-sensitive boundaries and performance risks). Cite the relevant requirement or rule and file location for each finding. Rank by user impact; distinguish verified defects from judgment calls. Do not repeat automated lint/test findings as manual style comments. If no spec exists, say so and review the stated request.

Use one agent by default. Return findings to implement for fixes and recheck; this specialist does not commit or mark a ticket done. Skip a review only when there is no code diff to inspect.

Sources: Matt Pocock's two-axis review and Addy Osmani's quality review.