Most AI agent skills are open-source recipes anyone can copy. That’s great for sharing but terrible for businesses with proprietary workflows. A new protocol aims to change that by keeping skills private while letting agents use them.
Explore how Open Agent SDK’s SessionStore safeguards conversation history and the four permission modes that control agent capabilities. Learn implementation details for secure persistent sessions.
A recent study uncovered alarming vulnerabilities in AI agent environments, including prompt injection and tool description poisoning, across widely used MCP servers. Researchers introduced an open-source scanner and a new vulnerability standard to address these risks.