[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f2lez6xrtuktkb":3},{"code":4,"message":5,"data":6,"timestamp":72},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":42,"summary_md":43,"topic":44,"github_topics":47,"license":68,"homepage":69,"updated_at":70,"history":71},"morluto\u002Frea",1,"Reverse engineer anything with agents, from app behavior down to native binaries.","TypeScript",377435,94316,88243,193,[16,22,28,33,38],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},50058850,"Reverse engineer anything with agents, from app behavior down to native binaries",100,38,"2026-10-12",{"id":23,"title":24,"points":25,"comments":26,"posted_at":27},50012930,"Reverse Engineer Anything",4,0,"2026-10-08",{"id":29,"title":30,"points":31,"comments":26,"posted_at":32},48901763,"Show HN: A MCP for Agents to reverse engineer binary code",3,"2026-07-14",{"id":34,"title":35,"points":36,"comments":36,"posted_at":37},50028294,"REA: Reverse Engineer Anything is now #1 trending repo on GitHub",2,"2026-10-10",{"id":39,"title":40,"points":8,"comments":26,"posted_at":41},49980896,"REA: Reverse Engineer Anything","2026-10-06","把本地二进制、JS\u002FElectron 与 .NET 分析工具接给编码 Agent，用来弄清某个功能到底怎么实现。","## 解决什么问题\n\n看到一个闭源应用里的功能想借鉴，或者想弄明白一个编译后的二进制到底做了什么，常规做法是打开 Hopper、Ghidra 或 IDA 啃汇编，再靠人脑把线索拼起来。编辑器里的 Agent 对此一无所知，知识被困在另一个工具里。REA 把逆向分析工具通过 MCP 暴露给 Agent，让它来发起调查，返回代码、引用和明确的未知项，再据此写出并测试一个类似实现。\n\n## 有什么不同\n\n传统反编译是交互式、单工具的：一个引擎、一个文件、一个人读输出。REA 的思路是让 Agent 来编排调查过程——它可以把已有的 Hopper、Ghidra 或 IDA 当作 provider，也可以完全不用原生引擎，退回静态 JavaScript\u002FElectron 分析；并且每条结论都绑定到产生它的证据上。这种“证据优先”的框架才是真正的差异点：工具输出是以带明确局限的发现形式返回，而不是一句自信的总结。同一套流程也提供 CLI，方便脱离 Agent 写脚本。README 偏重安装说明、方法细节不多，各个分析后端能做到多深，需要自己用了才知道。\n\n## 适合谁\n\n适合本来就熟悉反汇编、想让 Agent 帮忙干杂活的开发者和安全研究员：定位某个功能的代码路径、梳理 Electron 的 IPC 边界、快速摸一遍 .NET 程序集。如果你手头没有任何原生产分析工具、也不打算装，或者想要一个无需人工阅读的全自动审计，又或者主要面对高度混淆的恶意样本（这时工具的可信度比 Agent 的便利更重要），那它并不合适。README 提到移动端和固件目标，但都依赖额外的外部工具。\n\n## 上手成本\n\n需要 Node.js 和 npm，原生分析还要 Hopper、Ghidra 或 IDA 这类 provider。Linux 上跑 Ghidra、JADX 和固件工具时，`PATH` 里需要有 procps 兼容的 `ps`。安装命令是 `npx rea-agents setup`，它会改动你的 Agent 配置和工作流指令（有备份），装完必须人工核对并重启 Agent。项目迭代很快，官方要求你频繁更新，这本身就是维护成本。MIT 协议。近期发布说明反复强调只从官方渠道下载，说明山寨包确实是个风险，务必只从官方仓库、官网或 npm 包获取。\n\n## 点评\n\n想法很应景，MCP 这条链路也是真做出来了，但这个项目的数据很诡异：创建仅半年左右就有 376k stars、94k forks，周榜单周 +88k，却从没上过日榜。这种模式很少是自然增长，星标数字不宜当真。HN 上 100 分、38 条评论只能说明有人关注，证明不了正确性。如果你本来就配好了反编译工具，“Agent 与分析工具共享一座桥”这个思路值得一试；但要预期这是一个年轻、变化很快的代码库，而不是稳定的平台。",{"slug":45,"name":46},"security","安全",[48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67],"agent-skills","ai-agents","binary-analysis","claude-code","cli","codex","cordis","decompiler","developer-tools","dsh","dsh-plugin","ghidra","hopper","llm","mcp","model-context-protocol","rea","reverse-engineering","static-analysis","touhou","MIT","https:\u002F\u002Frea.tools","2026-10-24T14:09:12Z",[],1792870738]