Skip to content

Lab 3.5: AI Peer Review

Module: 3.5 - Multi-Claude Workflows | ← SlidesDuration: 30 minutes Sample Project: hackathon-starter

Learning Objectives

By the end of this lab, you will be able to:

  • Run multiple Claude Code sessions simultaneously
  • Use one Claude to write code, another to review
  • Work on parallel tasks in different terminals
  • Leverage multiple perspectives for better code quality

Prerequisites

  • Completed previous Day 3 labs
  • Ability to open multiple terminal windows
  • Familiarity with the hackathon-starter project

Why Multiple Claude Sessions?

Single Claude limitations:

  • One context, one perspective
  • Can't review its own work objectively
  • Sequential task execution

Multi-Claude benefits:

  • Fresh perspective for code review
  • Parallel task execution
  • Specialized roles (writer vs. reviewer)
  • Cross-checking for errors

Setup

bash
# Navigate to the sample project
cd sample-projects/hackathon-starter

# You'll need TWO terminal windows for this lab

Task 1: Writer-Reviewer Pattern

Time: 15 minutes

Use one Claude to write code, another to review it.

Terminal 1 - The Writer:

bash
claude

Give the Writer a task:

Create a new utility function in a file called utils/stringHelpers.js that:
1. Truncates a string to a max length with ellipsis
2. Converts a string to title case
3. Generates a URL-safe slug from a string

Include JSDoc comments for each function.

Wait for the Writer to complete.

Terminal 2 - The Reviewer:

bash
claude

Give the Reviewer a task:

Act as a senior developer performing a code review.

Review the file utils/stringHelpers.js and check for:
1. Edge cases that aren't handled
2. Performance issues
3. Naming clarity
4. Missing error handling
5. Test coverage suggestions

Be constructive but thorough. Rate the code 1-10 and explain.

Compare perspectives:

  • Did the Reviewer catch issues the Writer missed?
  • Were the suggestions actionable?

Success criteria:

  • [ ] Writer created the utility file
  • [ ] Reviewer provided meaningful feedback
  • [ ] At least one improvement identified

Task 2: Implement Review Feedback

Time: 10 minutes

Have the Writer fix issues found by the Reviewer.

Terminal 1 - Back to the Writer:

The code review found these issues:
[Paste the Reviewer's feedback here]

Please address each issue and update the file.

Terminal 2 - Re-review:

Re-review utils/stringHelpers.js after the fixes.
Did the developer address all the feedback?
What's the new rating?

Success criteria:

  • [ ] Writer addressed feedback
  • [ ] Reviewer confirmed improvements
  • [ ] Code quality improved measurably

Task 3: Parallel Task Execution

Time: 5 minutes

Work on two independent tasks simultaneously.

Terminal 1:

Add input validation to the user registration endpoint in controllers/user.js.
Validate email format and password strength.

Terminal 2:

Write unit tests for the existing user profile update functionality.
Focus on edge cases.

Both tasks run in parallel, saving time.

Success criteria:

  • [ ] Both tasks completed
  • [ ] No conflicts between changes
  • [ ] Time saved vs. sequential execution

Workflow Patterns

Pattern 1: Writer-Reviewer

Claude A (Writer) → Creates code
Claude B (Reviewer) → Reviews code
Claude A (Writer) → Fixes issues

Pattern 2: Parallel Workers

Claude A → Task 1 (e.g., frontend)
Claude B → Task 2 (e.g., backend)

Pattern 3: Devil's Advocate

Claude A → Proposes solution
Claude B → Challenges assumptions, finds flaws
Claude A → Defends or improves

Pattern 4: Specialist Teams

Claude A → Security-focused review
Claude B → Performance-focused review
Claude C → Readability-focused review

Tips for Success

  1. Name your sessions - Use /rename to track which Claude is which
  2. Clear context - Each Claude starts fresh, won't know what the other did
  3. Copy outputs - Manually share relevant info between sessions
  4. Use git worktrees - For larger parallel tasks, use separate worktrees

Git Worktrees for Parallel Work

For larger parallel efforts:

bash
# Create a worktree for the second Claude
git worktree add ../hackathon-feature-b feature-branch-b

# Terminal 1 works in main directory
# Terminal 2 works in ../hackathon-feature-b

Troubleshooting

Sessions conflicting:

  • Work on different files
  • Or use git worktrees for isolation

Context confusion:

  • Each session is independent
  • Explicitly share relevant context via copy-paste

Too many sessions:

  • Start with 2, add more only if needed
  • More isn't always better

When to Use Multi-Claude

ScenarioRecommendation
Quick code reviewYes - fresh perspective helps
Large refactoringMaybe - coordinate carefully
Parallel independent featuresYes - saves time
Debugging a single issueNo - one focused session better
Security auditYes - multiple reviewers catch more

Stretch Goals

If you finish early:

  1. Try the Devil's Advocate pattern on an architecture decision
  2. Set up git worktrees and work on two feature branches
  3. Create a "review checklist" command for the Reviewer Claude

Deliverables

At the end of this lab, you should have:

  1. Experience running parallel Claude sessions
  2. Code reviewed and improved by a second Claude
  3. Understanding of when multi-Claude helps
  4. Workflow patterns for your team

Next Steps

After completing this lab, move on to Lab 3.6: CI/CD & Headless Mode.

Claude for Coders Training Course