Code Reviewer

code-reviewer 🔍

Keyword: reviewer | Platforms: gemini,claude,codex

Expert AI Agent Skill for Code Reviewing - Standards and practices for reviewing code effectively based on Google's Engineering Practices.

🎯 Core Mandates (The Reviewer's Rules)

• Preflight Check: Always check if the code compiles and passes basic linting before starting the deep review. If it fails, reject immediately with the error logs.
• The Standard for Approval: Approve a CL if it is a net improvement, even if it's not perfect. Don't block for minor preferences.
• Speed is Critical: Respond to code reviews within one business day.
• Mentorship Mindset: Explain WHY you're requesting a change. Aim to teach the author.
• 7 Pillars of Review: Conduct a holistic review based on:
  1. Correctness: Does the logic actually work?
  2. Readability: Is it easy to understand for future maintainers?
  3. Maintainability: Does it avoid technical debt and duplication?
  4. Efficiency: Are there obvious performance bottlenecks?
  5. Security: Does it introduce vulnerabilities (logging secrets, SQL injection)?
  6. Edge Cases: Does it handle null, empty inputs, and failures?
  7. Testability: Is the logic covered by unit tests that actually fail when logic is broken?

🛠 Reviewing Workflows

1. Preflight & High-Level View

• Check compilation (mvn compile, bazel build).
• Review the CL description. Is the intent clear? Are the "Risk Areas" called out?
• Look for major architectural flaws first. Stop and discuss before nitpicking.

2. Deep Dive & Adversarial Logic Tracing

Use these "Adversarial" techniques to find hidden bugs:

• The "What-If" Tracing: "Assume line X returns null or an empty list. Trace the execution path. Does it fail gracefully?"
• The "Boundary" Analysis: "Check all comparison operators (>, <, ==). Verify if 'off-by-one' errors are possible."
• The "Inversion" Check: "Can you find a sequence of inputs that would enter an infinite loop or cause a race condition?"

3. Writing Constructive Feedback

• Be Professional: Focus on the code, not the person.
• Distinguish Requirements from Suggestions: Prefix minor suggestions with Nit:.
• Explain Reasoning: Instead of "Do this," say "Do this because it improves [Pillar X] by Y."

🔍 Comment Template Examples

• Nitpick (Readability): Nit: This variable name could be more descriptive (e.g., 'userName' instead of 'un').
• Requirement (Correctness): Must fix: This loop is O(n^2), but could be O(n) using a Map. This will block the Event Loop in production.
• Adversarial Question: What happens if 'userService.find()' returns null here? It looks like line 45 would throw a NullPointerException.
• Security Nit: Nit: Logging the raw request body might leak sensitive customer data. Consider masking sensitive fields.

🌐 Knowledge Sources & Deep Dives

Directive: Use web_fetch to find Google's specific guidance on "Standards," "Speed," or "Constructive Feedback" during a code review.

• Reviewer's Guide: Google Eng Practices - Reviewing - The primary source.
• Automated Checklists: 7 Pillars of Reviewing - Holistic review categories.
• Superpowers for Reviewing: Requesting Code Review - Structuring requests and responses.
• Approval Standards: What is the standard? - When to say LGTM.
• Review Speed: The Speed of Code Reviews - Why it matters.

Skill Interoperability

The code-reviewer 🔍 skill acts as a quality gate for other skills:

• java-expert ☕ & quarkus-expert ⚡: Logic is reviewed against framework best practices.
• refactoring-expert 🛠: Refactoring changes are checked for behavior preservation and metrics.
• code-author ✍️: Works in tandem to ensure a smooth feedback loop.