Two tools with minimal documentation. check_order_status has a basic description (56 chars) and simple schema with one integer parameter. search_knowledge_base has a longer description (168 chars) but vague parameter guidance. Neither tool documents output schema, error handling, or recovery paths. No parameter constraints, enums, or validation guidance. Descriptions lack WHEN/WHY context. No tool annotations (readOnlyHint present in metadata but not declared in schema). Both tools are READ_ONLY but this is not surfaced to the LLM via schema hints.

• Add explicit output schema to both tools. For check_order_status, document: {order_id: int, product: string, status: string, order_date: string, tracking_number: string|null}. For search_knowledge_base, document what the RAG chain returns (e.g., {answer: string, sources: [string], confidence: float}).
• Add parameter constraints: order_id must be a positive integer; question must be 5-500 characters. Include these in parameter descriptions, not just schema.
• Implement structured error responses. Instead of returning 'No order found with ID {order_id}' as a string, return {error: 'not_found', order_id: <value>, suggestion: 'Try searching with a different order ID or contact support'}.
• Add readOnlyHint: true to both tool schemas so LLMs know these are safe to call without confirmation.
• Expand search_knowledge_base description to explain: 'Searches a vector database of store policies, product info, and FAQs using semantic search. Returns the most relevant answer and source documents. Use when user asks about policies, shipping, returns, or products.'