Code & Chain · Signal Desk

Malicious Agent Skills Target Claude Code and OpenAI Codex, Evading Scanners

Original sourceCyber Security News

Summary

Researchers have uncovered malicious skill kits targeting Claude Code and OpenAI Codex, using structural obfuscation and self-extracting techniques to bypass eight common scanners, with hidden malicious code capable of stealing developer keys and crypto wallets.

Key points

  • Alerts developers to supply chain risks in AI coding tool skill marketplaces, providing specific defense recommendations.
  • AI coding skills become a new attack surface; malicious skills can bypass existing scanners, requiring heightened developer vigilance.
  • Developers should avoid auto-executing unknown skills, use sandbox behavior analysis, and manually review skill source code.
  • Similar to npm supply chain attacks, but this targets AI agent skill marketplaces, with a newer attack surface and stronger evasion capabilities.

Editorial note

This page is Code & Chain's editorial summary of public sources. It may be prepared with AI assistance and published through an automated workflow. Refer to the original sources; this content is not investment, legal, or tax advice.