# /reverse-engineer-anything: Reverse engineer shipped apps with evidence for every claim

Source: https://trendhub.net/skills/morluto-rea/reverse-engineer-anything
Skill folder: https://github.com/morluto/rea/tree/main/.agents/skills/reverse-engineer-anything
Pack: REA: Reverse Engineer Anything (https://trendhub.net/skills/morluto-rea)
Works with: Claude Code, Codex, Cursor, Gemini CLI and others, Terminal (global CLI)

The reverse-engineer-anything skill teaches an agent when and how to use REA's MCP tools on software it has no source for. It picks the first tool from the target type, works summary-first to save context, plans larger investigations as checklists of questions, and finishes by explaining findings in plain language with cited evidence and stated unknowns.

## When to use it

- You want to understand how a feature in a desktop, Electron or mobile app works and build something similar.
- You need to compare two versions of a shipped binary or package.
- You want to record a web app, live or from a saved HAR capture, as an agent training environment.

## How it works

1. **Check REA is needed**: For a complete source repository the agent uses normal tools; REA is reserved for claims that depend on shipped artifacts or runtime evidence.
2. **Route the target**: It picks the first tool by target type, such as JavaScript application analysis for an ASAR, a managed-assembly tool for .NET, or a native session for executables.
3. **Work summary-first**: It starts from default summaries and pages retained evidence only when a specific question remains, without repeating identical calls.
4. **Stage broad investigations**: It turns the request into a checklist, reuses prior sessions, follows each question from entry point to result and keeps a ledger of evidence and gaps.
5. **Report and close**: It explains findings tied to evidence IDs, separates observed behaviour from design choices when building, and closes or saves the session.

## What it steers the agent away from

- Claiming static analysis observed the program running
- Guessing which app the user meant instead of asking
- Re-running setup or diagnostics before every investigation

## What it can touch

Calls REA's local MCP tools to read binaries, packages and captures, can launch or attach to targets and browsers for requested runtime evidence, and may run npx rea-agents or rea doctor to repair a missing connection.

## Questions

### Do I need Ghidra or IDA?

Only for deep native analysis. Many built-in inspections, including JavaScript, .NET, Go and Rust metadata, PE tables and SQLite, need no native engine.

### What happens if the MCP connection drops mid-task?

The skill retries once only when the call is safe to repeat; otherwise it tells you and asks you to restart the MCP client.
