Anju4/rea-reverse-engineering-demo

REA Reverse Engineering Demo

A small, safe practice target for learning evidence-based reverse engineering with AI using REA (Reverse Engineer Anything).

This isn't a training dataset. It's a bundle of files for a hands-on exercise: a tiny Electron app with one mysterious Export invoices button, helper scripts, real REA output, and a rebuild-and-verify challenge.

It accompanies:

• Medium article: REA: The AI Tool That Can Help You Understand Software Without Its Source Code
• GitHub repo: Anjub004/ai-reverse-engineering-with-rea
• Kaggle notebook: Reverse Engineering with AI: Hands-on REA Demo

What's inside

Path	Contents
demo/invoice-exporter/	Tiny Electron app whose export rules aren't visible in the UI
demo/scripts/summarize-rea-output.js	Turns REA's large JSON evidence into a readable summary
demo/scripts/compare.js	Checks a rebuilt exporter against the original on 200 random invoice batches
demo/rebuild/my-exporter.js	A naive "guessed" exporter to fix
walkthrough/	Step-by-step exercise, answer key, and real REA output (rea-agents 6.7.0)
article/	The full article and its diagrams
docs/, prompts/	Setup guide and copy-paste prompts
PROJECT_README.md	The full project README

Quick start

Requires Node.js 22.19+, 24.11+ or 26+.

# download this dataset, then from its root:
npx -y rea-agents@latest analyze-javascript-application "$(pwd)/demo/invoice-exporter" --json > rea-output.json
node demo/scripts/summarize-rea-output.js rea-output.json
node demo/scripts/compare.js

Then follow walkthrough/README.md until compare.js reports 0 mismatches.

Responsible use

The demo app was written for this exercise and is MIT-licensed. Only analyse software you own or are authorised to examine. This is an independent learning project, not affiliated with the REA project.