Restaurant-agent has well-structured tool definitions with complete JSON schemas and clear descriptions. All 7 tools have proper input schemas with type constraints (enums, patterns, required fields). Descriptions are concise and role-annotated (viewer+, staff+, manager only). However, parameter descriptions are minimal or absent, most params lack explanatory text beyond their names. Output schemas are not documented, forcing LLMs to infer response structure. Error handling is basic (JSON error objects) but lacks recovery guidance. Tool composition is sound (single responsibility, clear naming), but missing parameter-level documentation and output schema documentation limits LLM reasoning quality.

• Add parameter descriptions to all tools. For search_menu_items: 'query: Free-text search term (e.g., "pasta", "vegetarian"). Matches item names and descriptions.' 'category: Filter by menu category (e.g., "appetizers", "mains"). Omit to search all categories.' 'excludeAllergens: Array of allergen codes to exclude (e.g., ["gluten", "dairy"]). See get_menu for full allergen legend.'
• Document output schemas in tool descriptions. E.g., get_menu: 'Returns {categories: [{id, name, items: [{id, name, description, price_eur, allergens}]}], allergen_legend: {code: description}}'. This enables LLMs to plan field extraction.
• Enhance error messages with recovery guidance. Instead of '{allowed: false, error: "permission_denied"}', return '{allowed: false, error: "permission_denied", reason: "Requires staff+ role. Current role: viewer. Try get_menu or search_menu_items instead."}'
• Add enum constraint to excludeAllergens parameter. Define valid allergen codes (gluten, dairy, nuts, shellfish, etc.) as an enum in the schema, or document them in the description with a reference to get_menu.
• Clarify date range constraints in get_sales_summary. Add to description: 'Date range must be within the past 2 years. Both dates are inclusive. Example: from=2024-01-01, to=2024-12-31 returns all sales in 2024.'