Three tools with complete input schemas and descriptions, but quality varies. get_weather and build_packing_list have detailed, LLM-optimized descriptions (120-180 chars) with well-typed parameters including optional fields with clear semantics. select_skill is simpler but adequate. All tools are READ_ONLY (good for safety). However, output schemas are not documented, LLMs cannot plan downstream calls or extract structured results. Parameter descriptions are good but lack explicit constraints (e.g., date format validation, skill enum values). No error handling guidance provided.

• Document output schemas for all three tools. For get_weather, specify: {forecast: [{date, temp_min, temp_max, condition, precipitation_mm}], location: {name, lat, lon}}. For build_packing_list, specify: {items: [{name, category, source}], total_count}.
• Convert select_skill 'name' parameter to a formal enum: {type: 'string', enum: ['cold_weather', 'beach']} instead of relying on description text.
• Add explicit date format guidance to get_weather and build_packing_list: 'ISO 8601 format (YYYY-MM-DD), e.g. 2025-01-15. Dates must be within 365 days.'
• Add error handling guidance: 'If geocoding fails, return available nearby locations. If skill not found, return list of valid skills. If date range exceeds 365 days, return error with max range.'
• Document the weather object structure expected by build_packing_list inline or via a reference schema, so LLMs understand what fields to extract from get_weather output.