Build a Codex-Powered AI Code Review Agent That Analyzes Diffs and Generates Structured Feedback

Build a fully functional web-based micro app that acts as an AI-powered Code Review Agent.

This app must allow a developer to paste:
• A Git diff
• A pull request snippet
• Or a full file

The AI agent analyzes the code and produces structured, actionable review feedback.

This is not a generic “improve this code” tool.
It must simulate a real senior engineer reviewing a PR.

⸻

Core Behavior

The agent must generate output structured into these sections:
1. Summary of Changes
2. Potential Bugs
3. Performance Concerns
4. Security Risks
5. Code Smells / Anti-patterns
6. Refactor Suggestions
7. Test Coverage Gaps
8. Overall Risk Level (Low / Medium / High)

Each section must contain clear reasoning and specific references to code patterns.

No vague statements.

⸻

Review Quality Rules

The AI must:
• Identify unused variables
• Detect possible async/await mistakes
• Flag improper error handling
• Suggest memoization where appropriate
• Detect unsafe patterns (eval, innerHTML misuse, unsanitized input)
• Recommend typing improvements if TypeScript is detected
• Identify duplicated logic
• Flag deeply nested conditionals
• Highlight potential race conditions

It must behave like a thoughtful reviewer, not a linter.

⸻

Unique Twist

Add a “Review Mode” selector:
• Strict → enterprise-grade review
• Balanced → pragmatic startup review
• Teaching Mode → detailed explanation of reasoning

Tone and verbosity must adapt based on selected mode.

⸻

UI Requirements

Dark mode only.
Premium developer dashboard aesthetic.

Layout:
• Left panel: code input area
• Right panel: structured review output
• Top: Review Mode selector + Analyze button
• Bottom: Export Review button (copy markdown format)

Include:
• Collapsible feedback sections
• Subtle severity indicators
• Loading state with progressive output rendering
• Review history stored locally

⸻

Technical Requirements
• Single-page app
• Pure HTML, CSS, JavaScript frontend
• Modular structure
• AI API integration layer (mockable)
• LocalStorage for review history
• Clean separation between:
• UI layer
• Review generation logic
• AI prompt builder

⸻

Internal AI Prompt Template (Used by App)

You are a senior software engineer performing a detailed code review.

Analyze the following code or diff carefully.

Provide feedback structured exactly into the following sections:

Summary of Changes
Potential Bugs
Performance Concerns
Security Risks
Code Smells / Anti-patterns
Refactor Suggestions
Test Coverage Gaps
Overall Risk Level

Be specific. Reference patterns in the code.
Avoid generic advice.
Match tone according to selected review mode.

Code to review:
{{PASTED_CODE}}

⸻

Output Requirements

Return:
1. Complete runnable frontend code
2. The internal AI prompt template
3. Explanation of how to connect to a real AI API key securely

No placeholders.
No pseudo-code.

⸻

Recommended LLMs

ChatGPT 5.2
Claude

⸻

Programming Languages

HTML
CSS
JavaScript

⸻

Tags

ai code review, codex automation, developer productivity, pull request review, vibecoding, ai agents, javascript tools, automation workflow

⸻

If you want, next we can build:
• An OpenClaw-style autonomous workflow agent
• An AI Daily Digest productivity agent
• A fully local task automation orchestrator
• Or something experimental and powerful

You choose the direction 🚀