What problem does it solve? Code reviews done by the same session that wrote the code are biased toward approval, and ad-hoc reviews miss architecture drift, missing tests, and security issues. This Skill runs every review in a fresh-context subagent that sees only the diff and committed docs, producing an unbiased, structured verdict posted directly on the GitHub PR. ## Core Features & Use Cases - Unbiased fresh-context review: Every invocation spawns a subagent with no session memory, so the verdict cannot be influenced by the conversation that produced the code. - Multi-dimensional scoring: Checks ADR and layer architecture alignment, per-language coding standards compliance, test coverage, a lightweight security scan, and acceptance criteria from the task board. - GitHub-native output: Posts one GitHub review carrying the summary verdict (APPROVED / CHANGES REQUESTED / BLOCKED) plus inline comments anchored to specific diff lines — no review files are written to the repo. - Use Case: Before merging a feature branch, ask for a review; the Skill resolves the PR's actual base branch, reads the relevant coding standards and ADRs, verifies visual evidence for UI changes, and posts an inline-annotated verdict on the PR. ## Quick Start Ask the assistant to review the current branch or a specific PR number before merging, for example by saying "review this PR before merge".