按分类精选 REA 使用案例
按用途实时整理的清单
以下清单来自本站已上线的案例,会随数据库更新,不是写死的仓库清单。各分类内按投票和 GitHub 星标数排序;这反映的是社区关注度,不是独立测试,也不代表本站或 REA 项目的背书。
从你想弄明白的事出发
选择最接近目标的分类:复现某个功能、梳理文件格式或通信协议、还原自己程序遗失的源代码、防御性安全工作、CTF 与学习,或围绕 REA 打造的工具。照着做之前先读案例的原始来源,并确认它使用的 REA 版本和分析引擎。
照着做之前
只在你拥有或获授权检查的软件上复现案例,并在隔离环境中处理不受信任的目标。每个收录的案例都通过了内容规范;空的分类只表示目前还没有收录。教学资源讲的是方法,不一定提供软件。
应用功能复现
- How Notion copies text and HTML — Traces a page-facing API in Notion's packaged JavaScript through its IPC wrappers to the system clipboard and shows how an HTML marker links rich block data to a copy.
- Untangle rebuilt from REA evidence only — An agent analysed only the shipped Windows binary of an open-source puzzle game, wrote its findings, froze them, rebuilt the game as a web page and was graded against the real source (52 of 54 claims correct).
- REA Research Lab: ten studies of released software — Ten bounded studies of released open-source components (Monaco, Grafana, n8n, Excalidraw, Joplin, Node-RED, KeePass, 7-Zip, restic, rclone), each with evidence, an independent reconstruction and thousands of differential checks.
补丁与修复
文件格式与协议
- ASIAIR firmware study with REA — Notes and scripts from studying how a telescope controller's firmware is built: its INDI-based stack, plain JSON-RPC on port 4700 and the three hardware generations; no vendor firmware is included.
- 一个蓝牙马桶的状态栏遥控器,来自其微信小程序 — 逆向工程了智能马桶微信小程序所使用的蓝牙协议,并用 REA 和 Codex 构建了一个菜单栏遥控器(展示页未链接原始 X 帖子)。
- QShare-linux:Linux 版快速共享 — 一个 Linux 上 Google 快速共享协议的开源实现,借助 AI 研究 Nearby 连接而构建,让 Linux 能向 Android 发送文件。
安全与恶意软件分析(防御)
- 对生产环境 Next.js 和 Stripe 应用的外部泄漏检查 — 对作者自己门户的黑盒审计:匿名访问者能看到什么,应用会泄露什么,使用 REA 从外部进行检查。
遗失源码还原
- Reconstructing DX-Ball 1.07 — Rebuilds the 1990s Breakout game's logic in C and checks it against the original x86 program with differential tests and byte-exact compiled output.
- TH04 source reconstruction — Reconstructs a PC-98 DOS shooter's code, matching its 16-bit routines to readable C++ and the historical compiler's output.
- The industrial revolution in reverse engineering — An essay on rebuilding three Touhou games with agents: what verification looks like, where agents help, and how knowledge carries from one project to the next.
- 我把自己寫的 iOS App 編成執行檔丟給 AI,不給原始碼。 用免費的 REA + Ghidra,16 分鐘把損益公式原樣還原出來。前兩次它老實說「不知道」 — 開發者將自己撰寫的 iOS App 編譯成執行檔後交給 AI,在未提供原始碼的情況下,以 REA 搭配 Ghidra 於 16 分鐘內還原出損益計算公式,並分享前兩次失敗到修正設定後成功的過程。
游戏模组与互通
- OpenBattlefield2:洁净室引擎重实现 — 对《战地2》背后 Refractor 2 引擎的 C++20 洁净室重实现,是引擎而非游戏;不重新分发任何游戏内容,也没有反作弊。
CTF 与学习
- 一个被剥离符号的 bzip2 构建:REA 返回了什么,并与源代码核对 — 向 REA 询问在一个被剥离符号、经过优化的 bzip2 二进制文件中,哪个函数打印了一条消息;其答案及 24 个被调用函数与未剥离符号的构建及源代码一致,文章列出了这并不能证明什么。
- OpenBSD netcat, analysed with REA and Ghidra — A static analysis write-up of Ubuntu's netcat-openbsd binary using REA with Ghidra, with the evidence behind each finding.
- rea-playbook: from evidence to file:line to recreation — A hands-on teaching playbook using a minimal Electron target: extract evidence with file and line references, probe the MCP server and recreate the feature.
- Hands-on: analysing your own apps with REA, Ghidra and Claude Code — Follow-along exercises from a Korean video: build your own small Electron app and C program, then analyse them with REA, Ghidra and Claude Code.
- How an authenticator generates a login code — Follows the six-digit code calculation of the open-source Aegis authenticator through its released Android APK, and uses a demo clock to show why the code changes every 30 seconds.
- Solving a CTF flag checker — Inspects a DownUnderCTF binary, finds its checking functions, turns the masks and sums into equations and confirms the solved flag against the original program.
工具与集成
- 在基于 Fedora 的 Linux 上运行 REA:设置说明与注意事项 — 在基于 Fedora 的 Linux 上运行 REA 的实用设置说明,包括过程中遇到的问题以及待测试队列。
- My local model setup for REA on Linux (one 12 GB GPU) — How to run REA with local models on a single 12 GB GPU, and how five models compared.
- 减少 REA 工作流中的词元浪费 — 一个展示与分享帖,讨论如何在 REA 工作流中降低词元使用量,以及其他人正在使用什么方法。
- REA (Reverse Engineer Anything) — Reverse engineer anything with agents, from app behavior down to native binaries.
- 基准测试:REA 在 Linux 上跨 Node 版本及五个本地模型的测试 — REA 在 Linux 上 Node 22、24 和 26 的启动计时,以及五个本地模型的对比。
- REA in depth: an MCP toolchain for source-free analysis — A Chinese-language technical survey of how REA lets a coding agent analyse software without its source code.