Artificial intelligence agent frameworks disagree on most design concepts, but all are adopting isolated sandboxes for code execution. Securing this execution boundary prevents malicious code from compromising host systems and should precede any framework choice.
FireBrief
FireBrief