Illustrated tutorial: take apart an app feature with REA, from observed behaviour down to implementation evidence图文教程:用 REA 拆解一个 App 的功能,从行为一路挖到实现证据
Run the full reverse-engineering loop with the REA that entered the library today: register it and back up your configuration with a single command, have the agent analyse a feature you care about and demand evidence and limitations, then ask it to build a similar feature in your own project.
用今天入库的 REA 走一遍完整逆向流程:一条命令完成智能体注册与配置备份,然后让智能体分析一个你关心的功能、要它给出证据与局限,最后让它在你自己的项目里实现一个类似功能。

Step 1: register with a single command第一步:一条命令完成注册
Make sure Node.js and npm are installed, then run npx rea-agents setup in your terminal.
Choose the agent you want to configure (Claude Code, Codex, Cursor, Gemini CLI, Grok Build and more). It shows the configuration changes it plans to make and backs up your existing configuration first.
Restart your agent afterwards. Native binary analysis needs an analysis engine: it can reuse an existing Hopper / Ghidra / IDA installation, and setup can install Hopper with your approval.
确认本机有 Node.js 与 npm,然后在终端执行 npx rea-agents setup。
按提示选择你要用的智能体(Claude Code、Codex、Cursor、Gemini CLI、Grok Build 等)。它会展示将要写入的配置改动,并在改动前备份原有配置。
确认后重启智能体。原生二进制分析需要分析引擎,可以复用本机已有的 Hopper / Ghidra / IDA,setup 也能在征得你同意后安装 Hopper。

Step 2: have the agent analyse a feature第二步:让智能体分析一个功能
State the goal in plain language and explicitly demand evidence, for example: "Understand how global search works in this Electron app, list the evidence, and tell me which parts are inference."
REA picks the right analysis tools (JavaScript static analysis needs no native engine — just hand it the unpacked directory or the ASAR file) and returns modules, import relationships and Electron boundaries together with the evidence.
A key habit: make it separate "observed behaviour" from "inferred implementation" and cite where the evidence comes from. Do not accept conclusions without provenance.
用自然语言给出目标并明确要求证据,例如:「帮我搞清楚这个 Electron 应用里的全局搜索是怎么实现的,把证据列出来,并说明哪些是推断」。
REA 会选择合适的分析工具(JavaScript 静态分析不需要原生引擎,直接把解包后的目录或 ASAR 交给它即可),跑完把模块、导入关系、Electron 边界等信息连同证据一起给你。
关键习惯:让它区分「观察到的行为」和「推断的实现」,并给出证据出处,不要接受没有出处的结论。

Step 3: review and archive in the terminal第三步:在终端复盘与存档
To inspect the raw result yourself, use the terminal: npx -y rea-agents@latest analyze-javascript-application /absolute/path/to/app --json. The output is structured JSON, easy to archive or paste into a report.
For regular use, install the CLI: npm install --global rea-agents, then run rea --help to see the native analysis commands and provider selection.
REA iterates quickly, so remember to run rea update periodically (Bun global installs need Bun 1.2.17 or later) and re-run setup as instructed afterwards to refresh the registration.
想自己看原始结果时,用终端命令:npx -y rea-agents@latest analyze-javascript-application /绝对路径/到/app --json,输出是结构化 JSON,方便存档或写进报告。
长期使用就装上命令行:npm install --global rea-agents,之后直接 rea --help 查看原生分析的命令与提供方选择。
REA 迭代很快,记得定期 rea update(Bun 全局安装需 Bun 1.2.17 以上),更新后按提示重跑一次 setup 刷新注册。
Step 4: turn the findings into your own feature第四步:把结论变成你自己的功能
Finally hand the requirement to the agent: "Based on that analysis, build a prototype of a similar feature in my project — propose a plan before writing code."
Because you already hold the evidence, this implementation is far more reliable than copying something by feel. Ask it to add tests once it is done.
最后把需求交给智能体:「参考刚才的分析,在我的项目里实现一个类似功能的原型,先给方案再写代码」。
因为前面已经拿到了证据,这一步的实现会比「凭感觉抄一个」靠谱得多;记得让它实现后补上测试。
Use reminder使用提醒
This article is compiled from public sources and ships with the matching resource. Product features and pricing are subject to the official page. Resources are for learning and exchange only — please respect the original license.
本文整理自公开资料并附上配套资源;涉及产品的功能与价格以官方页面为准。资源仅供学习交流,请遵循来源许可。
想马上用起来?去「AI工具」栏挑一个直接下载,或在「AI教程」里跟着图文步骤做一遍。
去 AI工具 → 看 AI教程 →阅读与点赞数据保存在你的浏览器本地,欢迎留下你的想法。
今日正能量学一点,用一点;今天种下的种子,会长成明天的能力。去免费下载专区 →广告
评论 文明发言,让讨论更有价值