Curious what standards exist for AI agent permissions. Something like Linux read, write, execute types, but for AI agents. For example, is there standard way to provide a "buy" permission, or agentic commerce agents?

I've seen a some stuff, but wanted to get community input.