New "SkillCloak" trick hides malware in Claude Code & Codex skills so scanners miss it: why it is not safe to use

What happened

On July 6, 2026, security researchers disclosed a technique called SkillCloak. It hides harmful code inside agent "skills," which are add-on packages that give AI coding tools like Claude Code and OpenAI Codex extra abilities. The team tested the trick against eight widely used security scanners and more than 1,600 real malicious skills. One method, which bundles code so it unpacks itself when it runs, fooled the scanners over 90 percent of the time. A second method fooled them over 80 percent. The hidden code still worked as intended once an agent loaded it. The research builds on a real attack called ClawHavoc, which planted hundreds of password-stealing skills on a public marketplace.

What it means for you

A skill runs with the same access as the agent that loads it. A booby-trapped one can reach your files, your saved logins, your keychain passwords, and your crypto wallets. Passing a security scan no longer proves a skill is safe.

What to do instead

• Install skills only from sources you already trust.
• Do not add a skill just because it passed a scanner.
• Favor skills a person has reviewed and tested, not only machine-checked ones.

AgentPod lists only skills our team has reviewed and tested by hand. That does not make any skill risk-free, but it adds a human check on top of automated scanning.

Sources

• https://cybersecuritynews.com/agent-skill-malware-targets-claude-code-and-openai-codex/
• https://thehackernews.com/2026/07/new-skillcloak-technique-lets-malicious.html