code-review-assist

Summarizes pull request diffs, surfaces review concerns, and assesses test coverage for human reviewers.

2.1k|283|Updated Mar 12, 2025
One-click install
npx skills add https://github.com/stacklok/toolhive --skill code-review-assist
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: code-review-assist
Source: https://github.com/stacklok/toolhive/tree/main/.claude/skills/code-review-assist
Command: npx skills add https://github.com/stacklok/toolhive --skill code-review-assist

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Reviewing pull requests thoroughly takes significant time, and reviewers often miss architectural concerns, weak test coverage, or leaky abstractions when scanning large diffs under time pressure.

Core Features & Use Cases

  • Structured PR Review: Produces a change summary, ownership/responsibility background map, categorized important changes, severity-labeled key concerns (blocker/suggestion/nitpick/question), and a testing assessment for each PR.
  • Review Session Planning: Discovers open and recently merged PRs via GitHub, estimates risk, detects stacked PRs, and tracks which PRs were reviewed, skipped, or deferred.
  • Continuous Improvement Tracking: Maintains a review-session-notes.md file comparing skill findings against the human reviewer's actual comments to refine future reviews.
  • Use Case: A senior engineer facing a queue of five PRs asks for a review session; the skill prioritizes them by risk, then walks through each one with concerns formatted as conventional comments ready to post.

Quick Start

Ask the assistant to start a code review session for my open GitHub pull requests, or provide a specific PR number to review in depth.

Frequently Asked Questions about code-review-assist

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I review a GitHub pull request with AI assistance?

Provide a PR number or ask for a review session, and the skill fetches the diff, summarizes the change, maps affected owners and responsibilities, and lists severity-labeled concerns. You stay in control of all final review decisions and posted comments.

How to prioritize multiple pull requests waiting for review?

The session planning step discovers open PRs requesting your review, recently merged PRs, and PRs with new pushes, then assigns each a high/medium/low risk estimate. You choose the review order: chronological, highest-risk-first, or by author.

What is the difference between this skill and an automated code reviewer?

This skill supports human-in-the-loop review sessions where you make the decisions, while autonomous reviewer agents check best practices independently. It prepares summaries, concerns, and draft comments but never posts anything without your approval.

Does it support conventional comments format for review feedback?

Yes, every drafted comment is prefixed with a conventional comments label: blocker, suggestion, nitpick, or question. Severity is calibrated aggressively, and all drafts are shown to you before posting.

Can it detect stacked pull requests and dependency chains?

Yes, during session planning it identifies PRs whose base branch is another PR in the list, groups them as a stack, and notes the dependency chain and each PR's responsibility within it.

When should I not rely on AI-assisted code review?

Do not treat its output as a final verdict; it explicitly avoids blanket approvals and states assumptions when context is missing. Security-critical or unfamiliar subsystems still require your direct inspection of the diff.