REA (Reverse Engineer Anything): the open-source tool that lets Codex and Claude Code investigate existing software

By Nate from Nate’s Substack · Thu, 15 Oct 2026 13:04:08 +0000.

The email introduces REA, short for Reverse Engineer Anything, an open-source tool that gives AI coding assistants such as Codex and Claude Code a way to investigate existing software, including programs whose source code isn't available. The user points their agent at an application or a supported browser session and asks how a specific feature works. REA provides the tools to inspect the code and its behavior, and the agent uses those findings to explain the mechanism and help build something of the user's own.

The author says he is looking at REA now because it has already passed 200,000 stars on GitHub and tops the platform's daily trending page, momentum he considers enough to examine in more detail what can actually be done with it.

The example case is a developer who used REA to investigate QuickBooks Desktop and then built a local accounting application with its own transaction and reporting engine. The project is called Ledger Reborn, and its author has published the research, including the comparisons used to test the new software.

The question Nate raises is what happens between looking inside QuickBooks and having a different application capable of doing useful work. His answer is that software development becomes far more accessible when an AI helps investigate an existing program: a feature used for years turns into a set of decisions, such as which records to keep, how to calculate a result or what should happen when something changes. Once those decisions are visible, it is easier to describe what your own version needs.

The author also points to a common problem: if you have ever asked an AI to build something and watched the result fall apart once you used it, this approach deserves attention, because a convincing screen tells you very little about what should happen next, what needs to be remembered or which figures must reconcile with each other.

In the accompanying video, Nate discussed the opportunity this creates for both developers and non-programmers. The full guide covers where to point REA, what to ask and what to do with the answer.

According to the email, the guide covers: (1) framing a question that can be finished, by naming a behavior and the decision it informs, so you know where to start and when to stop; (2) connecting and confirming REA, by installing it in Codex or Claude Code and running a small inspection that proves the tools loaded; (3) turning a business rule into a specification, using Ledger Reborn's research on QuickBooks to pin down a calculation and a check with sample data; (4) seeing what must survive a copy, by tracing how Notion's clipboard preserves block structure, to design copy and paste in your own editor; and (5) pointing REA at your own application, a team or a legacy program, with a starter prompt and a check that flags when the answer is wrong.

The author adds that the five investigations offer concrete starting points, each drawn from work published by the REA team or from a community account. The email is a preview: the full text, Nate's Library MCP and the Slack community are reserved for paid subscribers, so no further detail is available.