A prompt cookbook for teaching your AI agent to reverse engineer

What it does

This is a collection of prompts for REA (Reverse Engineer Anything), a toolset that lets AI coding agents inspect how software actually works — binaries, firmware, kernels, games, APKs, Electron apps, websites. Each prompt is a fill-in-the-blanks template: swap the <angle brackets> for your target and paste it into your agent. The repo covers feature tracing, native binary analysis, clean-room cloning, game logic recovery, WASM porting, firmware triage, kernel analysis, and vulnerability research.

The interesting bit

The real craft here isn’t the topics — it’s the discipline baked into every prompt. Each one forces the agent to separate confirmed facts from hypotheses, cite evidence (file, function, address), and flag what remains unknown. That’s the difference between an agent that reverse engineers and one that confidently hallucinates a call graph. The cloning section is also notably careful: it’s clean-room methodology, extracting behavior into a spec rather than copying code or assets.

Key highlights

• 16+ prompt categories spanning feature tracing, binary triage, protocol extraction, firmware, Android, kernels, and games
• Every prompt demands evidence-backed answers with confirmed/inferred/unknown separation
• Clean-room cloning workflow: black-box study → CLONE_SPEC.md → implement from spec only
• Dedicated WASM porting prompts, including a headless “logic core first” approach with state-hash comparison
• Includes a wildcard prompt and prompt-writing tips explaining why the templates work

Caveats

• It’s glue, in the best sense: the value is entirely in the prompt phrasing, and it depends on REA being installed and connected to your agent
• Several sections (anti-cheat, DMA, cracking, DRM) are legally sensitive; the README carries a prominent use-responsibly warning and restricts them to research, CTFs, bug bounties, and your own software
• The README leans on external sources (rea.tools, the official prompt gallery) for the underlying tooling

Verdict

Worth a bookmark if you already use REA or want your agent to explain how an unfamiliar binary behaves instead of guessing. If you don’t have an agent-plus-REA setup, this is a menu for a restaurant you can’t get into yet.