New 'SkillCloak' trick hides malware inside Claude Code and Codex skills that steal your passwords: why it is not safe to use

What happened

On July 9, 2026, researchers at the Hong Kong University of Science and Technology showed two ways to hide malware inside AI agent 'skills.' Skills are add-on packs that give tools like Claude Code and OpenAI Codex new abilities. The first method, called SkillCloak, wraps the malware so it unpacks itself only after you install it. Against 1,613 known bad skills, this let more than 90% of them dodge the security scanners meant to catch them. The best scanner fell from catching about 99% of threats to roughly 10%. The hidden code is tied to a campaign that planted over 300 poisoned skills on one marketplace. Install one, and it quietly steals saved browser logins, keychain passwords, SSH keys (the digital keys that log you into servers), and crypto wallets.

What it means for you

A skill runs with the same access as the agent that loads it. One marketplace listed over 40,000 skills within months, and a scanner's stamp of approval is no longer proof a skill is clean.

What to do instead

Install skills only from sources that review and test each one by hand, not just scan it. Stick to a small set you actually need. Check what a skill can reach before you run it. AgentPod lists only skills we have reviewed and tested ourselves, which lowers this risk but does not remove it, so stay cautious wherever you install.

Sources:

• https://www.helpnetsecurity.com/2026/07/09/malicious-ai-agent-skills-scan/
• https://thehackernews.com/2026/07/new-skillcloak-technique-lets-malicious.html