reverse-engineer-anything for Decompilation and Binary Analysis
機械翻訳 / Machine-translated

機械翻訳 / Machine-translated
Original article: reverse-engineer-anything for Decompilation and Binary Analysis (cccskills)
Skill page: reverse-engineer-anything
Key takeaways from this article
Installing reverse-engineer-anything lets your agent use REA to examine the contents of distributed apps and binaries.
It performs decompilation and runtime observation, then explains features and version-to-version differences with supporting evidence.
This skill is aimed at developers who need to investigate what's inside a distribution artifact.
This article walks through what can be analyzed, how to install the skill and set up REA itself, example prompts, and things to verify before you start.
You receive a report on a distributed app's functionality with evidence and unknowns clearly separated.
For example, if you have an Electron app's ASAR file, you might write:
Analyze this ASAR and explain how settings are saved.
The agent first identifies the type of target and, for an ASAR, selects the analysis approach suited to JavaScript apps.
The results include an evidence record called Evidence and a graph showing how code is connected.
The limits of the analysis and any unknowns are listed alongside the findings.
Each conclusion in the explanation is linked back to its Evidence entry.
Observed facts, inferences drawn from them, and points that remain unclear are distinguished from one another, so the reader can judge how much to trust each part.
When results are large, start with the summary and then drill into only the sections that address your remaining questions.
This skill teaches the agent how to use the REA suite of analysis tools.
When source code alone cannot explain the behavior of a distribution artifact, the agent searches for answers through decompilation, inspection of bundled files, and runtime observation.
Primary uses are explaining how features work, comparing versions, and gathering materials for reconstruction.
The analysis method used first depends on the target.
The table below summarizes representative targets and the conditions required for each.
| Target | Examples | Requirements |
|---|---|---|
| Electron / JavaScript | ASAR, extracted JavaScript tree | No MCP registration or native analysis engine needed when analyzing via CLI |
| Archives / packages | ZIP, APK, IPA, MSIX, AppX, DMG, PyInstaller executables | Obtain a file listing and hashes |
| Android | APK code, classes, methods | Android analysis tools |
| Managed assemblies | .NET and other PE/CLI assemblies | Dedicated analysis tools |
| Firmware | Firmware images | User-supplied Binwalk or Unblob on Linux |
| Native code | .NET NativeAOT PE/ELF, etc. | Ghidra analysis to obtain pseudocode |
| Recorded traffic | HAR, SAZ, packet captures | Packet captures require Linux and TShark 4.x |
Beyond the table, the skill also supports apps bundled into a single executable with Bun or Deno, and resources from Apple apps.
Embedded information in Go and Rust binaries, as well as SQLite databases, are also supported.
An audit function is available to check whether credential traces or information that should not be public remain in a distribution artifact.
There is also a feature to generate an environment bundle for agents (AgentEnv bundle) from web apps and Linux desktop apps.
On the other hand, REA is not used for ordinary structural analysis of a complete source repository.
In that case, the agent uses the usual repository tools.
There are two installation methods: running a command yourself, or asking the agent to do it.
The command is shown below; when you run it you can choose the target environment such as Claude Code or Codex.
npx skills add https://github.com/morluto/rea --skill reverse-engineer-anything
If you prefer to ask the agent, send the following message as-is:
次の Agent Skill をインストールしてください。
https://github.com/morluto/rea/tree/main/.agents/skills/reverse-engineer-anything
まず SKILL.md と付属ファイルの中身を確認し、不審な処理(外部への送信、認証情報の読み取りなど)がないことを確かめてから、Claude Code なら ~/.claude/skills/reverse-engineer-anything/、Codex なら ~/.agents/skills/reverse-engineer-anything/ に配置してください。
The installation location depends on whether you want to use the skill across all projects or only in a specific project.
| Agent | Use across all projects | Use in a specific project only |
|---|---|---|
| Claude Code | ~/.claude/skills/ | .claude/skills/ in the project |
| Codex | ~/.agents/skills/ | .agents/skills/ in the project |
Installing the skill alone does not register the REA MCP server or install any analysis engines.
The exception is when you install it as a plugin from the Claude Code, Codex, or Cursor Marketplace.
In that case the skill is bundled with the plugin, and the plugin also registers the REA server.
In all other cases, register with rea setup; if something isn't working, use rea doctor to check the status.
Before making any configuration changes, the target path and a backup are shown and user approval is requested.
If you only need to analyze an ASAR or an extracted JavaScript tree, you can run analysis via the CLI below without registering MCP.
When running it, specify an absolute path from a trusted directory outside the target project.
npx -y rea-agents@latest analyze-javascript-application /absolute/path/to/app --json
Including both the target you want to examine and what you want to know in your prompt helps convey your intent clearly.
Here are three examples.
Compare two versions of a distributed app and organize the differences with evidence and unknowns.
The differences between the two versions are returned as an explanation linked to the evidence record.
Observed differences, inferred differences, and points that could not be confirmed are written separately, making it usable as a changelog.
Analyze this APK and explain the classes and methods involved in the login process.
The agent examines the APK's code and identifies the relevant classes and methods along with where they are called, with supporting evidence.
Note that when you only want a list of files contained in the archive, an archive-level analysis is selected instead.
I want to recreate this app — please identify and document with evidence the code, configuration, and resources inside the package.
For this request, the agent obtains a listing and hashes of the package's contents and organizes the materials needed for reconstruction.
When the listing is large, a summary is returned first; specify the parts you need and drill down from there.
The external tools and OS requirements you need to supply separately vary depending on the analysis target.
rea setup does not install or update Node.js, npm, Homebrew, Java, Ghidra, IDA, JADX, Binwalk, Unblob, or similar tools.
Firmware and packet-capture analysis assume Linux, so verify the requirements for your target in advance.
Static analysis and runtime observation are treated as distinct.
Analysis that only opens and reads files cannot confirm actual runtime behavior.
Runtime observation may involve launching the app, making connections, making changes, and cleanup, so make sure you understand what will happen before requesting it.
It is also important not to treat the unknowns and inferences listed in a report as established facts.
Audit results for a distribution artifact are not a guarantee that the app is safe.
When using analysis findings to make decisions, anchor your reasoning in the parts labeled as observed facts in Evidence.
The skill itself is licensed under MIT.
Before analyzing anything, verify the terms of service and rights associated with the target app yourself.
Reading through SKILL.md and the accompanying files (the guides in the references folder) before installing is also a good idea.
No.
The skill only provides instructions; you also need to register the REA MCP server and set up an analysis engine separately.
For ASARs and JavaScript trees, you can analyze without registering MCP by using the CLI.
REA is not used for ordinary structural analysis of a complete source repository.
It is suited for cases where source code alone cannot explain a distribution artifact's behavior, or when you want to compare source against a distributed artifact.
Yes.
On Linux, it uses user-supplied Binwalk or Unblob to examine the regions.
When extracting contents, specify a new output directory using an absolute path.
reverse-engineer-anything is a skill for using REA to examine the contents of distributed apps and binaries with supporting evidence.
It proceeds with analysis using the method suited to the target — ASAR, APK, package, firmware, and more.
Results are returned with observed facts, inferences, and unknowns clearly separated, making it useful for version comparison and gathering reconstruction materials.
Alongside installing the skill, prepare REA itself and any required external tools, verify the rights associated with your target, and then give it a try.
cccskills page
This article is a repost of the cccskills article "reverse-engineer-anything for Decompilation and Binary Analysis."
Details about the skill and installation instructions can be found on the cccskills reverse-engineer-anything page.