Why you should not install unvetted AI agent skills

An audit of public AI agent skill registries in early 2026 found hundreds of skills that installed malware or exfiltrated credentials, according to a report from AISkills402. In the researchers' own test, the instruction that compromised a Claude agent was not an overtly malicious command but a step that appeared to belong to the task, underscoring that agent skills execute with whatever permissions the user has granted. The report recommends reading every file in a skill, scanning for known malicious patterns, and running it first with tools disabled.

A skill is a file your agent obeys with your permissions. What went wrong in public registries, the trap that fooled our agent, and how to check one first. An agent skill is a text file that your agent reads as instructions and then carries out with whatever access you gave it: your files, your terminal, your keys, sometimes a wallet. That is the whole risk. Researchers who audited public skill registries early in 2026 found hundreds of skills that installed malware or sent credentials away, and in our own test the instruction that fooled a Claude agent was never the one that shouted; it was a step that seemed to belong to the task. In a skill, every step seems to belong to the task. Before you install an unvetted skill, read every file in it, search it for the patterns below, and run it first with tools switched off. Read the full report on AISkills402: https://aiskills402.com/blog/agent-skills-security-risks https://aiskills402.com/blog/agent-skills-security-risks