Single tool with minimal definition quality. Tool name 'get_forecast' follows verb_noun convention (+). Description is present but generic at 34 chars, below the 50-200 char LLM-optimized range. Input schema has proper types (float) and basic descriptions, but lacks critical metadata: no output schema documented, no error handling guidance, no pagination support despite calling external API, no timeout/retry semantics. Tool returns raw API response (dict) without shaping for agent consumption. No parameter constraints (latitude/longitude ranges), no validation rules, no recovery guidance on API failures.

• Expand tool description to 80-150 chars: 'Get weather forecast for a location. Returns current conditions and 7-day forecast. Requires valid latitude (-90 to 90) and longitude (-180 to 180).'
• Add parameter constraints in schema: latitude type=number, minimum=-90, maximum=90; longitude type=number, minimum=-180, maximum=180. Update descriptions to include ranges.
• Document output schema explicitly. Specify returned fields (e.g., current_temperature, forecast_daily, timezone) with types and descriptions so LLM knows what to extract.
• Wrap external API call with try-except. Return structured error: {'error': 'API unavailable', 'retry': true, 'message': 'open-meteo service temporarily down. Try again in 30s.'} instead of raw exception.
• Shape API response before returning. Extract only essential fields (temperature, precipitation, wind_speed, forecast_days) and convert to markdown or structured summary. Reduce token waste from verbose metadata.