Does Claude Code CLI modify code during a review?

No. Claude-based reviews are read-only and do not apply changes to the codebase.

What if Claude Code CLI isn't installed or accessible?

Ensure Claude Code CLI is installed and available in your PATH; verify with claude --version.

How should I act on the findings from a Claude review?

Review the issues by severity, inspect the referenced file paths and lines, and implement fixes or add PR notes as appropriate.

claude-review

npx machina-cli add skill dceoy/ai-coding-agent-skills/claude-review --openclaw

Files (1)

SKILL.md

2.7 KB

Claude Review Skill

Use Claude Code CLI to perform automated code reviews that identify issues and suggest improvements. This is a read-only analysis skill.

When to Use

User asks to "review" code
User wants to check for bugs or issues
User mentions "security", "performance", or "quality"
Before committing code
During pull request review
User asks "what's wrong with this code?"

Prerequisites

Verify Claude Code CLI is available:

claude --version  # Should display installed version

Basic Usage

Step 1: Determine Scope

Decide what to review:

Uncommitted changes
Specific file(s)
Last commit
Entire codebase

Step 2: Check Current State

git status
git diff --stat
git diff

Step 3: Execute Claude Code Review

Run Claude Code with a review-focused prompt (use -p/--print to output once and exit):

claude -p "Perform comprehensive code review of [SCOPE].

Check for:
1. CRITICAL ISSUES (must fix): security, runtime errors, data loss, breaking changes
2. IMPORTANT ISSUES (should fix): logic bugs, performance, error handling, type safety
3. SUGGESTIONS (consider): refactors, readability, docs

For each issue:
- Severity (Critical/Important/Suggestion)
- File path and line number
- Why it's a problem
- How to fix it

Do NOT make any changes - this is review only."

If your CLI does not support -p, run claude and paste the prompt.

Step 4: Present Findings

Organize results by severity and include file references.

Example Reviews

claude -p "Review all uncommitted changes for bugs, security issues, and performance problems. Do NOT modify code."

claude -p "Security review of src/auth/*.ts focusing on SQL injection, XSS, auth/authorization, secrets, and input validation. Do NOT modify code."

Best Practices

Categorize by severity
Include specific file paths and line numbers
Explain why each issue matters
Verify findings to avoid false positives

Error Handling

If Claude Code CLI is not available, ensure it's installed and in your PATH
If the review is too broad, narrow the scope
If issues are unclear, request clarification or smaller scope

Related Skills

claude-ask for understanding code before reviewing
claude-exec for fixing issues found in review

Limitations

Static analysis only
May generate false positives
Cannot run tests or execute code

Source

git clone https://github.com/dceoy/ai-coding-agent-skills/blob/main/skills/claude-review/SKILL.mdView on GitHub

Overview

Use Claude Code CLI to perform automated, read-only code reviews that identify bugs, security vulnerabilities, performance issues, and quality problems. This helps reviewers catch issues before commits or PRs and speeds up feedback.

How This Skill Works

It runs Claude Code CLI with a review-focused prompt to enumerate issues by severity, capture file paths and line numbers, and provide explanations and suggested fixes. It never modifies source code; you review the findings and decide on actions.

When to Use It

User asks to review code
User wants to check for bugs or issues
User mentions security, performance, or quality concerns
Before committing code
During pull request review
User asks 'what's wrong with this code?'

Quick Start

Step 1: Determine scope (uncommitted changes, specific file(s), last commit, or entire codebase).
Step 2: Run Claude Code with a review-focused prompt (use `-p`/`--print` to output once and exit): claude -p "Perform comprehensive code review of [SCOPE]. Check for: 1. CRITICAL ISSUES (must fix): security, runtime errors, data loss, breaking changes 2. IMPORTANT ISSUES (should fix): logic bugs, performance, error handling, type safety 3. SUGGESTIONS (consider): refactors, readability, docs. For each issue: - Severity (Critical/Important/Suggestion) - File path and line number - Why it's a problem - How to fix it Do NOT make any changes - this is review only."
Step 3: Present findings organized by severity with file references and actionable next steps.

Best Practices

Categorize findings by severity (Critical/Important/Suggestion)
Include precise file paths and line numbers
Explain why each issue matters and its potential impact
Verify findings to avoid false positives
Keep the review read-only and provide clear next steps

Example Use Cases

claude -p "Review all uncommitted changes for bugs, security issues, and performance problems. Do NOT modify code."
claude -p "Security review of src/auth/*.ts focusing on SQL injection, XSS, auth/authorization, secrets, and input validation. Do NOT modify code."
claude -p "Review changes before committing to catch critical issues."
claude -p "PR review: analyze new code in reducers for potential runtime errors."
claude -p "Full codebase review for quality issues without modifying code."

Frequently Asked Questions

Add this skill to your agents