Two tools with basic verb-noun naming (get_alerts, get_forecast) and READ_ONLY risk classification. Both have input schemas with typed parameters and descriptions. However, descriptions are minimal (under 50 chars each), lack context on when to use each tool, and do not explain output structure. No output schemas documented. Error handling returns plain strings with no recovery guidance. Parameters lack constraints (e.g., state code validation, lat/lon bounds). Tool composition is reasonable but descriptions fail to distinguish use cases or guide LLM selection.

• Expand descriptions to 100 - 200 characters. Example: 'Get active weather alerts for a US state (e.g., tornado warnings, flood advisories). Returns a list of alerts with event type, severity, and instructions. Use this to check for hazardous conditions in a region.'
• Add output schema documentation. Specify that get_alerts returns a formatted string of alert objects with fields: event, area, severity, description, instructions. get_forecast returns periods with temperature, wind, and detailed forecast text.
• Add parameter constraints: state as enum of valid 2-letter US codes (CA, NY, TX, etc.); latitude as number with min -90, max 90; longitude as number with min -180, max 180.
• Replace plain error strings with structured error responses. Example: 'State code invalid. Must be 2-letter US code (e.g., CA, NY). Got: "XX". Try search_states() to find valid codes.' Include retryable flag.
• Add a discovery tool (e.g., list_states or describe_weather_tools) to help LLMs understand when to use each tool and what state codes are valid.