An agent has a search_tickets tool that returns every matching ticket as a JSON blob with all fields, including full comment histories, and identifies each ticket and assignee by a raw UUID. A broad query often returns tens of thousands of tokens, the agent's context fills up, and it sometimes passes the wrong UUID to a follow-up update_ticket call. Which redesign best addresses both problems?