A pass for every tool call: secure-by-default tools for AI agents (scope, approval, prompt-injection and exfiltration guards, credential injection, audit) with permission checks answered live by the systems they act in.