PlugWiki

Code Review Excellence Skill for Claude Code: A Four-Phase PR Review Guide

Gives your assistant a four-phase process for reviewing pull requests and writing feedback that is specific, kind and prioritized.

Installs
30.6K
on skills.sh
Stars
42.1K
on GitHub
Files
2
besides SKILL.md
Updated
Oct 2026
last repo commit

This skill gives your assistant a method for reviewing code. It works through four phases: gather context, review the design, go through each file line by line, then summarize and decide.

It also shapes the tone of the feedback with good and bad examples, a comment template and a severity marking. Two reference files cover review techniques and language-specific patterns, and SKILL.md tells the agent when to read each.

Good for

  • Developers reviewing pull requests who want a consistent method
  • Tech leads setting review standards or mentoring juniors
  • Anyone who wants review comments that teach instead of nitpick

Install code-review-excellence

Claude Code

  1. 1

    Copy this prompt.

    Install the code-review-excellence skill in Claude Code. Run this command:

npx skills add https://github.com/wshobson/agents --skill code-review-excellence

Then check that it loaded and tell me how to use it.
  2. 2

    Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

From its listing on skills.sh. Source.

What's in the skill

When it loads

Master effective code review practices to provide constructive feedback, catch bugs early, and foster knowledge sharing while maintaining team morale. Use when reviewing pull requests, establishing review standards, or mentoring developers.

Sections in SKILL.md

  1. When to Use This Skill
  2. Core Principles
  3. 1. The Review Mindset
  4. 2. Effective Feedback
  5. 3. Review Scope
  6. Review Process
  7. Phase 1: Context Gathering (2-3 minutes)
  8. Phase 2: High-Level Review (5-10 minutes)
  9. Phase 3: Line-by-Line Review (10-20 minutes)
  10. Phase 4: Summary & Decision (2-3 minutes)
  11. Giving Difficult Feedback
  12. Pattern: The Sandwich Method (Modified)
  13. Handling Disagreements
  14. Best Practices
  15. Common Pitfalls
  16. Templates

Key facts

Author
Seth Hobson (GitHub: wshobson)
License
MIT
Installs on skills.sh
30,592 (Nov 2, 2026)
GitHub stars
42,081 for the whole repo (Nov 2, 2026)
What installs
3 files including SKILL.md: references/review-techniques.md and references/review-patterns.md
Part of
The developer-essentials plugin, which also holds 10 other skills and 1 agent
Skill-only install
The README says the skills CLI or gh skill installs one skill without the plugin's agents, commands or hooks
Last commit
Oct 30, 2026 (repo)

What it includes

  • Goals and non-goals of a review

    Goals: catch bugs, keep code maintainable, share knowledge, enforce standards, improve design, build team culture. Not goals: showing off, nitpicking formatting, blocking progress or rewriting to taste.

    Try: “Review this pull request using the code-review-excellence skill.”

    Review mindset

  • Specific, educational, code-focused, balanced and prioritized

    Includes bad and good comment pairs, such as replacing "This is wrong." with a named race condition and a suggested mutex.

    Effective feedback

  • What to review by hand and what to leave to tools

    Review logic, security, performance, tests, error handling, docs, API design and architecture. Leave formatting, import order, lint violations and simple typos to tools.

    Review scope

  • Read the description, PR size, CI status and requirement

    Check whether the PR is over 400 lines and should be split, and whether tests pass.

    Phase 1: Context (2 to 3 min)

  • Architecture, file organization and testing strategy

    Ask whether the solution fits, whether a simpler way exists and whether tests cover edge cases.

    Phase 2: High-level review (5 to 10 min)

  • Logic, security, performance and maintainability per file

    Checks edge cases, off-by-one errors, input validation, injection risks, N+1 queries, leaks, names and magic numbers.

    Phase 3: Line-by-line (10 to 20 min)

  • Summarize concerns, note what was good, then approve, comment or request changes

    Offer to pair on complex changes.

    Phase 4: Summary and decision (2 to 3 min)

  • Read before writing feedback

    Checklist method (security, performance, testing), the question approach, suggest-don't-command wording and severity levels.

    Read · Reference: review-techniques.md

  • Read before language, architecture, test or security review

    Python and TypeScript or JavaScript patterns, architectural and test-quality review, and a security checklist.

    Read · Reference: review-patterns.md

  • Context, specific issue, helpful solution; handling disagreement

    Replaces the praise-criticism-praise sandwich. When the author disagrees: seek to understand, acknowledge valid points, provide data, escalate if needed, and know when to let go.

    Giving difficult feedback

  • Eight habits such as reviewing within 24 hours

    Limit PRs to 200 to 400 lines, review in blocks of at most 60 minutes, use tools, automate what you can, build rapport and be available to pair.

    Best practices

  • Seven mistakes such as rubber stamping and ghosting

    Perfectionism, scope creep, inconsistency, delayed reviews, ghosting, rubber stamping and bike shedding.

    Common pitfalls

From the official docs

Questions

What does the code-review-excellence skill do?

It gives your assistant a four-phase method for reviewing pull requests: context, design, line by line, then a summary and decision. It also covers how to word feedback and how to handle disagreement.

How do I install it?

Use the install box above. It installs only this skill. The wshobson/agents README also describes a plugin route, but the developer-essentials plugin adds 10 more skills and an agent.

Does it check out the pull request or run anything?

No scripts are bundled and SKILL.md names no commands. It is a method for the assistant to follow when it reads code you give it.

Does it check security and performance?

Yes, as checklist items in the line-by-line phase and in the review-patterns reference. They include input validation, injection, XSS, N+1 queries and memory leaks. It is a prompt to look, not a scanner.

Will it post comments on GitHub?

SKILL.md describes a comment template and a verdict but does not mention posting to GitHub. Paste or post the review yourself.

Is it safe?

It is guidance text plus two reference files, and SKILL.md does not tell the agent to run anything. Review its comments before you share them, as with any AI feedback.

Does it work in the Claude app?

The repo documents no Claude app route. The README covers Claude Code, Codex, Cursor and skills-only installs.

Information

updated Nov 2, 2026

Repository
wshobson/agents
Folder
plugins/developer-essentials/skills/code-review-excellence

wshobson/agentsskills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 7
  1. code-review-excellence SKILL.md · github.com
  2. wshobson/agents README · github.com
  3. developer-essentials plugin.json · github.com
  4. wshobson/agents LICENSE · github.com
  5. code-review-excellence on skills.sh · skills.sh
  6. review-techniques.md · github.com
  7. review-patterns.md · github.com

Updated Nov 2, 2026.