pbakaus/code-review-assistant

Code Review Assistant

Analyzes a pull request to assign reviewers, detect code issues, and visually explain complex changes. Can automatically assign reviewers via GitHub CLI and post review comments.

Setup: Run bash scripts/setup.sh to check environment and install dependencies.

Get PR Data

If you have a diff and PR link provided: Analyze that directly.

Otherwise: Use GitHub CLI to fetch from current repo:

gh pr view --json number,title,body,author,files
gh pr diff

If gh not installed: https://cli.github.com/

Analysis

1. Assign Reviewers: Read reference/team-expertise.md, match expertise areas to changes in the diff

2. Detect Issues:

   • Read reference/code-standards-map.md for quick issue list
   • If any patterns match the diff, grep the full details from code-standards.md using the XML tag
   • Flag by severity (🔴🟡🔵)

3. Visual Explanations: Unless none or only minor code changes, run:

   • node scripts/generate-diagram.js "<description>"
   • node scripts/upload-image.js <file> → Cloudinary URL
   • Include as ![](url) in review

Output Format

## 🤖 Claude PR Assessment

## 🎨 Visual Explanation
![](url) - Description (don't include if minimal code changes)

## 👥 Recommended Reviewers
- **@username** (Name) - Expertise areas that match this PR

## ⚠️ Areas of Concern
- 🔴/🟡/🔵 **Matched code standard pattern title**
  - Files it occurs in, optional 3-5 line snippets and aggregated explanation
  - Concise recommended solution

## ✅ Overall Assessment
Strengths, changes needed, recommendation

Post-Review Actions

After presenting the review, offer to help with:

1. Auto-assign reviewers:

   gh pr edit <number> --add-reviewer username1,username2,username3
   

   Use the GitHub usernames (@username) from the recommended reviewers list.

2. Post review as PR comment:

   gh pr comment <number> --body-file <review-file.md>
   

   Save the full review markdown to a temporary file first, then post it.

Ask the user if they'd like you to perform either of these actions.

Reference Files

Progressive disclosure - only load what you need:

• reference/team-expertise.md - Team member expertise areas with GitHub usernames
• reference/code-standards-map.md - Quick list of detectable issues (check this first)
• reference/code-standards.md - Full issue details with XML tags (grep specific sections)

Scripts

• scripts/generate-diagram.js - Gemini API integration for visual explanations
• scripts/upload-image.js - Cloudinary upload for public image URLs

Run with Node.js, see script headers for usage details.