Leases, Not Switches: Rethinking Permission Models for AI Agents

Most AI agent frameworks model permissions as switches: can-read-files yes/no, granted once,...