review

Run a pre-landing review on the output of git diff origin/main or paste the diff into chat. The review flags string interpolation in SQL queries as a critical finding with file and line references, and recommends parameterized queries instead.

It runs two passes: a critical pass covering SQL and data safety, race conditions, LLM output trust boundaries, and enum completeness, then an informational pass covering dead code, test gaps, crypto entropy, time windows, type coercion, and view performance.

Yes. The skill accepts either a git diff against the base branch or a diff pasted directly into the chat, applying the same checklist categories and severity classification to both inputs.

It uses a fix-first heuristic: mechanical fixes like dead code removal, N+1 eager loading, and magic number extraction are applied automatically, while security issues, race conditions, and large fixes are batched into a single question for the user.

It fetches Greptile line-level and top-level comments via the gh API, classifies each as valid, already fixed, false positive, or suppressed based on per-project history, and posts evidence-backed replies using tiered templates.

The design checklist only runs when the diff touches frontend files, detected via a diff-scope helper. It flags AI-slop patterns, typography issues, missing focus states, and violations of the project's DESIGN.md, and skips silently otherwise.