What problem does it solve? AI-assisted code changes often ship with unstated assumptions, needless complexity, sprawling scope, and no verification. This Skill inspects a concrete git diff before commit or merge and returns a short, actionable verdict instead of a style lecture. ## Core Features & Use Cases - Four-Principle Review: Surfaces hidden assumptions, flags over-engineering, checks that every hunk is required by the task, and demands a verifiable goal such as a test or observable output. - Regression Scope Analysis: Traces import fan-out, maps changed files to existing tests, flags schema and migration risks, and produces a three-tier minimum verification suite via the absorbed regression-scope-analysis reference. - Clear Verdict Output: Reports blast radius (files, hunks, net lines), per-principle findings with file:line, and a keep / keep with nits / fix first verdict with numbered blocking fixes. - Use Case: Before opening a PR for an AI-generated feature, run the review on git diff --staged to catch a smuggled refactor, a missing test for new behavior, and an unvalidated nullability assumption. ## Quick Start Ask the AI to review my staged diff with the diff-review skill and give me a keep or fix-first verdict before I commit.