code-reviewer

Paste one diff or name one file and ask for a code review. The skill confirms scope, classifies the path as backend, frontend, or repo-wide, applies the matching rule set, and returns findings ranked Blocker, Major, Minor, or Suggestion with suggested fixes.

Reviews check the repository pattern with scoped DI, soft delete via RecordStatus kept off wire DTOs, migrations for schema changes, connection strings read from appsettings.json, a global exception handler, and integration tests using TestWebApplicationFactory with per-class in-memory SQLite.

No. The skill is strictly advisory and read-only: it explains findings and suggests fixes but never edits, creates, deletes, or commits files, and it does not run tests or linters unless explicitly asked to verify a fix.

Yes. It applies a manual WCAG checklist covering labels, keyboard operation, focus styles, contrast, and live regions, and it escalates reasoning effort for a11y-heavy UI such as forms, modals, and tables. Context7 lookups verify WCAG techniques when uncertain.

When invoked by the CI script, it returns only a JSON array of findings with severity (Blocker, Major, or Minor), a description, and a repo-relative filePath or null. Interactive reviews instead use a Markdown template with summary, findings, and positive notes.

Use them only when uncertain about current library behavior or WCAG guidance, such as React 19 hooks, EF Core patterns, Tailwind v4, or Radix APIs. The limit is three queries per review, and queries must never include file contents, secrets, or PII.