Dev Agent Code Reviewer

A multi-agent AI review system designed to simulate how real software engineering teams review code from different perspectives.

This repository demonstrates how specialized AI agents can collaborate to analyze:

• maintainability
• security
• testing confidence
• engineering quality

Instead of relying on one generic assistant, the review process is divided into focused agents with clear responsibilities.

The goal is not only to generate feedback.

The goal is to improve engineering reasoning.

---

Why This Project Exists

Most AI-assisted development workflows focus heavily on code generation.

This repository explores another approach:

AI as a technical reviewer

The project demonstrates how specialized agents can help developers:

• detect risks earlier
• improve maintainability
• reduce regressions
• improve testing quality
• identify unsafe assumptions
• improve review consistency

The workflow is inspired by how real software teams review pull requests collaboratively.

---

Multi-Agent Review Workflow

The repository uses a specialized multi-agent workflow.

Each agent reviews the same implementation from a different perspective.

Implementation
      ↓
Code Reviewer Agent
      ↓
Security Reviewer Agent
      ↓
Test Reviewer Agent
      ↓
Consolidated Review
      ↓
Refactor Proposal

The final review combines findings from all agents into one actionable report.

---

Agents

1. Code Reviewer Agent

Focuses on:

• clean code
• maintainability
• architecture
• separation of concerns
• naming quality
• TypeScript usage
• refactoring opportunities
• long-term scalability

Main question:

Is this implementation understandable and maintainable over time?

---

2. Security Reviewer Agent

Focuses on:

• input validation
• unsafe assumptions
• trust boundaries
• sensitive data exposure
• authorization risks
• insecure identifier generation
• unsafe error handling

Main question:

Could this implementation create practical security risks?

---

3. Test Reviewer Agent

Focuses on:

• missing tests
• edge cases
• regression risks
• invalid inputs
• testability
• confidence before merge

Main question:

Do we have enough confidence to safely evolve this code?

---

Repository Structure

.github/
  copilot-instructions.md
  agents/
    code-reviewer.agent.md
    security-reviewer.agent.md
    test-reviewer.agent.md

docs/
  agent-purpose.md
  review-checklist.md
  workflow.md
  multi-agent-workflow.md

examples/
  bad-code-example.ts
  improved-code-example.ts
  review-output-example.md

prompts/
  code-reviewer-prompts.md
  security-reviewer-prompts.md
  test-reviewer-prompts.md
  consolidated-review-prompts.md
  refactor-prompts.md
  youtube-demo-prompts.md

---

Example Workflow

1. Review bad implementation

Example:

examples/bad-code-example.ts

Run:

• Code Reviewer Agent
• Security Reviewer Agent
• Test Reviewer Agent

---

2. Consolidate findings

Generate:

• critical issues
• maintainability risks
• security risks
• regression risks
• suggested improvements

---

3. Refactor implementation

Apply:

• safer validation
• explicit typing
• improved maintainability
• safer error handling
• better testing strategy

---

Educational Goal

This repository is intentionally educational.

The examples are designed to:

• stay simple enough for demos
• remain realistic enough for meaningful reviews
• demonstrate agent specialization clearly
• show practical engineering reasoning

The objective is not to replace human reviewers.

The objective is to help developers think more critically about software quality.

---

Recommended Use Cases

This repository can be adapted for:

• pull request reviews
• engineering mentoring
• AI-assisted code reviews
• onboarding examples
• review workflows
• developer education
• multi-agent experiments
• prompt engineering demos

---

Future Improvements

Possible future extensions:

• orchestrator agents
• automated review pipelines
• CI integration
• agent memory
• review scoring
• architecture reviewers
• performance reviewers
• observability reviewers

---

Final Idea

The most important idea behind this repository is:

Different engineering problems require different ways of thinking.

That is why the workflow separates:

• maintainability
• security
• testing

into specialized AI reviewers instead of using one generic assistant.