deobfuscating-malicious-javascript

Deobfuscates malicious JavaScript from droppers, web pages, and HTA/scriptlets: unrolling string-array and char-code encodings, resolving eval/Function chains, and statically recovering payloads and U

SKILL.md

GitHub

Deobfuscating Malicious JavaScript

When to Use

  • You have obfuscated JS (from a phishing page, HTA, .js dropper, or scriptlet) and need its real behavior.
  • You must recover hidden URLs, dropped commands, or a next-stage payload.
  • You want to statically unroll common obfuscation rather than run untrusted code.

Do not use a real browser or node to execute the script for analysis — that runs the malware. Use a sandboxed interpreter (box-js) or static transformation only.

Prerequisites

  • A safe analysis approach: static decoding, or a JS malware sandbox (box-js) in the lab.
  • Familiarity with common obfuscation: string arrays, String.fromCharCode, hex/unicode escapes, eval/Function chains, packers.

Safety & Handling

  • Treat the script as live code; never execute it outside an isolated sandbox.
  • Defang recovered URLs and store dropped payloads password-protected.

Workflow

Step 1: Normalize and de-pack

Pretty-print the source and identify the obfuscation style (array-shuffle, eval-packer, charcode). Decode static encodings first: hex/unicode escapes and base64 literals.

python scripts/analyst.py decode dropper.js

Step 2: Unroll string arrays and char-code builds

Reconstruct strings built from arrays/fromCharCode/concatenation to reveal API names, URLs, and commands.

Step 3: Resolve eval/Function indirection — safely

Replace eval/new Function with logging (or a sandbox) so the constructed code is captured as data, not executed, then recurse on the recovered layer.

Step 4: Extract behavior and IOCs

Identify the dropper actions (WScript.Shell, ActiveX, fetch/XHR), recover URLs and dropped paths, defang, and route payloads onward.

Validation

  • Static encodings (hex/unicode/base64) are fully decoded.
  • String-array/charcode constructions are unrolled to readable strings.
  • eval/Function layers are captured as data and recursed, with no untrusted execution.

Pitfalls

  • Running the script to "see what it does" and infecting the analysis host.
  • Stopping at one layer when the dropper nests several.
  • Missing environment-keyed branches (only acts on certain dates/locales) during static review.

References