Skip to main content

Claude Code Review Guide: How to Review AI-Generated Code

· 4 min read
Felo AI
Felo AI
Operations

Complete guide to reviewing AI-generated code. What to check, common issues, using /review command, and building a review workflow.

Code review is one of the highest-leverage activities in software development. It catches bugs before production, enforces standards, and spreads knowledge across the team. But it's also time-consuming and inconsistent. Claude Code can make code review faster and more consistent. The key: it needs to know your project's standards, not just generic best practices. Set up context-aware code review with MemClaw → memclaw.me

The Problem with Generic AI Code Review​

Claude Code review — generic syntax-only review vs context-aware architecture review comparison

Most developers who try AI code review are disappointed. They paste code into Claude and ask for a review. They get:

  • Generic best practices they already know
  • Suggestions that contradict their architecture
  • "Consider using X" where X is something they already rejected
  • Missing the project-specific issues that actually matter The problem isn't Claude's capability — it's context. Claude doesn't know your project, so it can't review against your standards.

Context-Aware Code Review​

The solution: give Claude your project context before reviewing. Without context: You: Review this code. Claude: Consider using GraphQL for more flexible queries... [You use REST only — this is useless] With MemClaw context: You: /review Claude: Loading project standards from MemClaw workspace...

Critical​

src/routes/orders.ts:47 Direct DB query in route handler. Your architecture requires Repository Pattern (/lib/repositories/). → Move to OrderRepository.findByUser()

Warning​

src/components/Cart.tsx:23 JWT stored in localStorage. Your decision (2026-02-15): httpOnly cookies only. → Use existing useAuth() hook The second review is useful because Claude knows your actual standards.

Setting Up Context-Aware Code Review​

Claude Code review — persistent project workspaces for AI agents

Step 1: Add your standards to MemClaw​

In your MemClaw workspace, add a code review standards section:

Code Review Standards​

Architecture Rules​

  • All DB access via Repository Pattern — no direct queries in routes
  • JWT in httpOnly cookies — never localStorage
  • React Query for all async operations
  • Error boundaries at route level, not component level

Naming Conventions​

  • Components: PascalCase
  • Hooks: useXxx
  • Utilities: camelCase
  • Constants: UPPER_SNAKE_CASE

Review Priorities​

  • Security: SQL injection, XSS, CSRF, hardcoded credentials
  • Performance: N+1 queries, missing indexes, unnecessary re-renders
  • Error handling: Unhandled promises, missing error boundaries
  • Architecture violations: Direct DB in routes, business logic in components

Step 2: Create the /review slash command​

.claude/commands/review.md: Review the staged changes in this repository.

First, load project standards from the MemClaw workspace.

Check for:

  1. Security issues (SQL injection, XSS, CSRF, hardcoded credentials, missing input validation)
  2. Architecture violations (check workspace for our patterns)
  3. Performance issues (N+1 queries, missing indexes, unnecessary re-renders)
  4. Missing error handling (unhandled promises, missing error boundaries)
  5. Test coverage gaps
  6. Anything that contradicts past decisions (check workspace)

Output format:

Critical (must fix before merging)​

Warning (should fix, explain why)​

Suggestion (optional improvements)​

Each issue: file:line, description, suggested fix.

Step 3: Run the review​

git add . claude /review Claude loads your project standards from MemClaw and reviews the staged changes against them.

What Context-Aware Review Catches​

Claude Code review — workspace dashboard with project standards

Architecture violations: "This route queries the database directly. Your convention is to use the Repository Layer in /lib/repositories/." Convention violations: "This component uses useEffect + fetch. Your convention is React Query for all async operations." Past decision conflicts: "We decided on 2026-03-10 not to use Redis caching in this module due to race condition issues. This implementation reintroduces the same pattern." Project-specific security issues: "This endpoint doesn't check the idempotency key. You have a known issue with Stripe webhooks firing twice — this needs an idempotency check." None of these appear in generic code review. They require project context.

Keeping Review Standards Current​

Your review standards should evolve with your project. When you make a new architectural decision, add it to the workspace: "Add to workspace: we decided to use optimistic locking for inventory updates. Any code that modifies inventory without optimistic locking should be flagged in code review." Next time you run /review, Claude knows this new standard.

Getting Started​

  1. Add your code review standards to your MemClaw workspace
  2. Create the /review slash command
  3. Run git add . && claude then /review before your next commit Set up context-aware code review → memclaw.me