---

name: gh-pr-review description: Review GitHub pull requests using the gh CLI. Use when a user asks for a PR review, architectural feedback, or wants a structured checklist and summary based on gh pr view/diff/comments or gh api pull request comments.

GH PR Review

Overview

Perform a structured PR review using gh commands and a consistent checklist. Provide concise findings and questions tailored to the PR scope.

Workflow

1. Identify the repo and PR number

   • Accept a PR URL or explicit owner/repo + PR number.
   • Ask only if missing.

2. Fetch PR metadata

   • gh pr view <num> --repo owner/repo
   • Use title/body, labels, reviewers, and stats to frame scope.

3. Fetch the diff

   • gh pr diff <num> --repo owner/repo
   • If diff is large, focus on critical files or API/DB changes first.

4. Fetch discussion context (optional)

   • gh pr view <num> --repo owner/repo --comments
   • gh api repos/owner/repo/pulls/<num>/comments for line threads.
   • Use when the user asks about a specific discussion or rationale.

5. Apply the standard review checklist

   • Purpose/Scope: Confirm intent and changes match the stated goal.
   • Correctness: Identify logic errors, edge cases, or state inconsistencies.
   • Domain/Model: Check invariants, naming, and aggregate boundaries.
   • Data/DB: Validate constraints, NULL rules, FK integrity, migrations.
   • API/Compatibility: Confirm request/response changes and version impact.
   • Performance: Watch for heavy queries, missing indexes, N+1.
   • Ops/Observability: Consider batch timing, retries, logging, metrics.
   • Tests: Ensure critical paths are covered; note missing tests.

6. Respond with a structured review

   • Prioritize issues by severity.
   • Cite file/line when available; otherwise point to section/topic.
   • Ask targeted questions when assumptions are unclear.
   • If the user asked for guidance only, do not post comments.

Output Guidance

Follow the platform review style. Keep summaries short and lead with findings. If the PR is a design-doc only, focus on architecture, data model, and risks.