Agent skills
Also called: Agent Skills · Claude skills
Agent skills are packaged instructions, scripts and resources that an AI agent loads to perform a task, such as Anthropic's Agent Skills for Claude. Because a skill tells the agent what to do and can include code, a malicious or tampered skill acts like a malicious package with the agent's access.
How it works
Skills are shared through marketplaces, repos and organisations. A skill can hide instructions in its text, bundle scripts that run with the agent's permissions, or pull dependencies at runtime. Scanners that review skills can themselves be evaded.
How to defend
Install skills only from sources you trust, review their instructions and bundled code, and run agents that use third-party skills with limited access.
Real cases 2 stories
HIGH
Supply chain · AIR Security
OpenAI Sesame: an official skill sent app tokens to an unowned domainOpenAI Sesame is a flaw in render-deploy, an official OpenAI skill that told coding agents Render private services sit at render-internal.com, a domain Render did not own. AIR Security registered it and received live bearer tokens.
MED
Attack · AIR Security
Anthropic skill scanner bypassed: malicious skills marked safeAIR Security researchers say Anthropic's skill and plugin scanner for Claude organization skills marked malicious skills as safe, using look-alike domains, an unclaimed PyPI dependency and an obfuscated binary.