Several agents need consistent logging, authorization, and exception handling. What should the implementation evaluate?

1. Hide exceptions so every task appears successful
2. Copy unrelated ad hoc checks into every prompt only
3. Reusable middleware or shared orchestration components with explicit contracts
4. Remove permission checks from specialist tools