code-review-excellence

Follow a phased process: gather context from the PR description and CI status, review architecture and test strategy at a high level, then check line-by-line for logic, security, and performance issues. Finish with a summary grouped by severity and a clear approve or request-changes decision.

Check logic correctness and edge cases, security vulnerabilities like injection and missing input validation, performance issues like N+1 queries, test coverage, and maintainability. Skip formatting and linting since automated tools handle those.

Use severity labels like blocking, important, and nit so authors know what must change. Phrase issues as questions or suggestions with rationale and code examples rather than commands, and balance criticism with praise for good work.

Watch for mutable default arguments, overly broad except clauses, and mutable class attributes shared across instances. These are frequent Python bugs that cause shared state and swallowed exceptions.

Skip manual review effort for formatting, import ordering, and lint violations, which automated tools handle. Also avoid this process when there are no code changes, when the discussion is design-only, or when the task is implementing fixes rather than reviewing.