Weather Forecast

Goal

Fetch weather forecast data from Open-Meteo using location and time parameters inferred from conversation context and the current user prompt.

Always produce deterministic output artifacts and avoid ambiguous interpretation.

Workflow

1. •

   Resolve user intent into request parameters.

   • •Required: location.
   • •Optional: time window.
   • •If location is missing, ask one concise follow-up question.
   • •If time is missing, default to "today" in the resolved location timezone.
   • •Convert relative time terms (for example: "tomorrow", "next Friday") to absolute dates before calling APIs.
2. •

   Resolve location with Open-Meteo geocoding first.

   • •Use Open-Meteo geocoding API to convert location text to coordinates.
   • •Do not use generic web search for coordinates unless Open-Meteo geocoding fails.
   • •If multiple geocoding matches are plausible, ask one disambiguation question (for example: "Springfield, IL or Springfield, MA?").
   • •If no geocoding result is found, ask for country/state context or direct coordinates.
3. •

   Build a valid forecast query.

   • •Use Open-Meteo forecast API as the primary weather source.
   • •Select hourly or daily fields based on request scope:
     • •Intra-day or hour-specific requests: include hourly fields.
     • •Day-level requests: include daily fields.
   • •Validate start/end dates:
     • •If only one date is given, use it for both start and end.
     • •If start is after end, ask for correction.
4. •

   Generate deterministic output.

   • •Always use uv for Python-related operations.
   • •Prefer Python standard library for request orchestration and file handling.
   • •Write one JSON artifact to output/weather-forecast.
   • •If the directory does not exist, create it using Python standard library (os).
   • •Use a unique filename format:
     • •weather_<location-slug>_<start-date>_<uuid8>.json
   • •Slugify location for filename safety (lowercase letters, numbers, dashes only).
5. •

   Return both a concise summary and structured data artifact.

   • •Provide a short human-readable summary in the assistant response.
   • •Include artifact path in the response.
   • •JSON output must include this shape:
     • •request: original request fields inferred from prompt/context.
     • •resolved_location: name, latitude, longitude, country, timezone.
     • •forecast_window: start_date, end_date, timezone.
     • •units: temperature/wind/precipitation units used.
     • •data: hourly and/or daily forecast arrays returned by API.
     • •source: Open-Meteo endpoint URLs used.
     • •generated_at: ISO-8601 timestamp in UTC.
6. •

   Handle failures explicitly.

   • •On transient network/API errors, retry once.
   • •If still failing, return a concise error with the failed step and missing requirement.
   • •Never fabricate weather values.
7. •

   Clean up temporary files only.

   • •Remove temporary scripts or intermediate files created for the task.
   • •Do not delete final artifacts in output/weather-forecast.