Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already signal destructiveness, but the description adds crucial nuance: default deletion is recoverable via trash, while permanent: true is immediate and irreversible. This goes beyond the scalar hints and gives the agent the safety-relevant behavioral context it needs.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.