Two tools with basic verb-noun naming (get_alerts, get_forecast). Both have descriptions (60-80 chars, below the 194-char baseline) and visible input schemas with typed parameters. However, schemas lack output documentation, parameter descriptions are minimal, and error handling is silent (returns plain strings on failure with no guidance for LLM recovery). No pagination, no structured output, no error classification. Descriptions are functional but sparse, they state WHAT but not WHEN or WHY to use each tool.

• Add output schema documentation to both tools. Define return type as structured object with fields (e.g., alerts: [{event, area, severity, description}], total_count, next_cursor) instead of plain strings.
• Expand parameter descriptions: 'state' → 'US state abbreviation (2 letters, e.g., CA, NY, TX)'; 'latitude'/'longitude' → 'Decimal degrees, range -90 to 90 and -180 to 180 respectively'.
• Replace silent error handling with actionable messages. E.g., 'Unable to fetch alerts for state CA: API returned 404. Verify state code is valid (2-letter abbreviation).' Include retry guidance.
• Enhance tool descriptions to 150 - 250 chars. Add WHEN to use: 'Call this to check for active weather warnings (tornado, flood, etc.) in a specific US state. Use before get_forecast to alert users to hazards.'
• Add pagination to get_alerts: accept optional limit (1 - 100, default 20) and offset parameters; return total_count and next_offset in response.