Safe runtime for autonomous on-chain AI agents: isolated sandboxes, Library skills, encrypted secrets.
