Obfuscated Payload Detection

What it is

Obfuscated payloads are malicious code that is deliberately encoded or wrapped to avoid detection. In the ClawHavoc campaign, attackers used base64 encoding with eval() — for example, eval(atob('BASE64_STRING')) — to hide the AMOS infostealer payload inside skill scripts. File-hash scanners fail because the hash changes with each re-encoding, and keyword scanners miss the payload because the malicious content is not visible in plaintext.

How TrustSkills detects it

TrustSkills checks skill code for obfuscation patterns including eval(atob()), exec(base64.decode()), Function() constructor calls with string arguments, and multi-level encoding chains. The scanner also detects staged payloads — code that downloads additional JavaScript or Python from a remote URL at runtime rather than including the malicious content in the skill package itself.

What we check

• eval(atob()) and exec(base64.decode()) patterns in JavaScript and Python skill code
• Function() constructor calls with string arguments (Function('malicious code'))
• Multi-level encoding chains (base64 inside base64, hex encoding of base64)
• Dynamic require() or import() of URLs rather than package names
• fetch() or axios calls to URLs ending in .js, .py, or .sh at runtime

Deep dive