What changed in this Electron update? Diffing two app.asar builds without running either

Your Electron app's 1.5.0 is ready. Can a web page in it now reach something it couldn't in 1.4.0? Did a window lose its sandbox? Does the code call a server it didn't call before?

The diff of two minified app.asar bundles won't tell you. Running the new build to find out is exactly what you wanted to avoid. So I built a tool that reads both builds statically and tells you what changed, with the file and line behind every change, and an honest list of what it couldn't figure out.

Everything below is reproducible in about ten seconds. You need Node 22+.

1. Two builds, one command

npx hexvero demo

This copies two builds of a small example Electron app (NotesDesk 1.4.0 and 1.5.0, written for these examples, nothing from a real product) into ./hexvero-demo and compares them.

2. What a page can reach

In Electron, a page talks to the main process through the preload script: contextBridge exposes members, each sends on an IPC channel, and ipcMain.handle answers. hexvero joins the three:

interfaces (2)
  + preload.js::notesdesk::checkUpdate  @ b: preload.js:9, handler main.js:30 · link direct
  - preload.js::notesdesk::openInShell  @ a: preload.js:9, handler main.js:30 · link direct

1.4.0 let any page open a path in the system shell through openInShell. 1.5.0 removes it and adds checkUpdate, whose handler at main.js:30 fetches an update URL. link direct means the bridge member and the handler use the same literal channel, so the link isn't a guess.

3. What a window may do

windows (1)
  ~ main.js::win  @ b: main.js:8
      nodeIntegration: true → false, nodeIntegrationInEffect: "on" → "off", sandbox: false → true

Settings are reported as written. If a value isn't written and rests on Electron's default for the version in package.json, it says default on / default off instead of pretending it read it.

4. What it could not tell

This is the part I care about most:

Not established
  ! b: 1 IPC call(s) pass a channel that is not a literal or a constant that could be followed
    (a parameter, a computed value). … ipcMain.handle at main.js:33. (ipc_channel_unresolved)

1.5.0 registers plugin:${name} channels with names from an environment variable. No static reading can list them. So instead of guessing (or silently dropping them), hexvero names the call and the line, and the IPC finding turns from "confirmed" to "inferred". Same in comparisons: an item in a file the other build didn't read is unknown, never removed.

5. Check it, write it down

npx hexvero ui --demo --open
npx hexvero compare hexvero-demo/notesdesk-1.4.0.asar hexvero-demo/notesdesk-1.5.0.asar --report report.md

The local web UI (127.0.0.1, per-session token, nothing loaded from the network) opens the source lines behind every file:line. The report is plain Markdown you can attach to a release review; same inputs, same bytes. Here's the example report.

What it doesn't do (honestly)

• It reads files; it never runs the app. A finding is what the code contains, not what it does, and not proof that anything is exploitable.
• It misses things. On my newest blind test set (an app written by someone without access to hexvero, measured once) it found 12 of 57 IPC channels and 5 of 8 version changes, mostly because channels came from a registry function it doesn't evaluate. The misses are published in the repo.
• "Doesn't execute the target" isn't a safety guarantee: parsing hostile bytes is still parsing hostile bytes (bounded in size, depth and time).

On your own app

npx hexvero compare old/MyApp.app/Contents/Resources/app.asar \
                    new/MyApp.app/Contents/Resources/app.asar --report report.md

It also reads ELF/PE/Mach-O, APKs, .NET, WebAssembly and more, and ships an MCP server so a coding agent (Claude Code, Cursor, Codex…) can run the same investigation with small, cited answers.

It's MIT and open source: github.com/maximilianfeix/hexvero. If it gets a fact wrong on your app, an issue with the file and line is the most useful thing you can send me. And if it showed you something you didn't know about your build, a star helps others find it.

I'm the author of hexvero. This post was written by an AI agent working on my behalf; every output in it was produced by running the tool. Only use it on software you own or are allowed to examine.