weather-plugin

Guide for using the weather plugin — data queries, radar imagery, model maps, and meteorological calculations

SKILL.md

GitHub

Weather Plugin Usage Guide

You have 12 weather tools. Prefer lightweight data tools and only use images when asked.

Tool Selection

"How's the weather in [place]?" → wx_conditions "What's the forecast?" → wx_forecast "Any warnings/watches?" → wx_alerts "Show me the radar" → wx_radar_image "Show me CAPE/temperature/wind map" → wx_model_image (use comma-separated vars: "cape,helicity,uh,srh" for multiple in ONE call) "What's the severe weather outlook?" → wx_severe "Weather in Tokyo/London/etc" → wx_global (worldwide) "Calculate dewpoint/wind chill/LCL" → wx_calc "Is there a tornado threat?" → wx_severe first, then wx_model_image --var cape or --var helicity for visual "What does the HRRR show?" → wx_model_image with appropriate var "Show me the 18z HRRR" → wx_model_image with cycle: 18 "Sounding at a point" → wx_sounding

IMPORTANT: Only these 12 tools exist

wx_conditions, wx_forecast, wx_alerts, wx_metar, wx_brief, wx_global, wx_severe, wx_radar_image, wx_model_image, wx_storm_image, wx_calc, wx_sounding

Do NOT try to call any other tool names.

Rules

  1. Start lightweight. Use wx_conditions or wx_brief before heavier tools.
  2. Images only when asked. Don't call image tools unless the user wants to SEE something.
  3. US vs international. Use wx_global for non-US locations.
  4. Calculations are verified. wx_calc uses metrust — 205 Rust-backed functions verified against MetPy.
  5. Don't stack redundant calls. wx_brief already includes conditions + forecast + alerts.
  6. BATCH model images. NEVER call wx_model_image multiple times for different variables. Use comma-separated vars in ONE call: "cape,helicity,uh,srh". This generates all images at once.
  7. Severe weather package. For severe weather analysis, use: var: "cape,helicity,uh,srh" — one call, 4 images.
  8. ZOOM to threat area. When the user asks about a specific region (e.g., "Indiana to New York"), ALWAYS pass lat/lon and radius_km to zoom in. Don't show full CONUS when they asked about a corridor. Examples:
    • "Ohio Valley" → lat: 39.5, lon: -83, radius_km: 400
    • "Indiana to New York" → lat: 40.5, lon: -77, radius_km: 500
    • "Texas" → lat: 31.5, lon: -99, radius_km: 500
    • "Southeast" → lat: 33, lon: -85, radius_km: 600
    • "Great Plains" → lat: 38, lon: -99, radius_km: 600 Use your geographic knowledge to pick appropriate center + radius.

wx_calc Examples

  • Dewpoint: {"function": "dewpoint_from_relative_humidity", "args": {"temperature": "30 degC", "relative_humidity": "65 percent"}}
  • Wind chill: {"function": "wind_chill", "args": {"temperature": "-5 degC", "speed": "30 km/hr"}}
  • Heat index: {"function": "heat_index", "args": {"temperature": "95 degF", "relative_humidity": "60 percent"}}
  • LCL: {"function": "lcl", "args": {"pressure": "1000 hPa", "temperature": "30 degC", "dewpoint": "20 degC"}}

Image Tools

wx_radar_image and wx_model_image return PNG file paths in JSON.

For radar: provide NEXRAD site ID (KRTX=Portland, KTLX=OKC, KLOT=Chicago) or lat/lon. Default 1024px, 200km range, 10dBZ noise filter.

For model images: specify variable (cape, refl, temp, dewpoint, rh, gust, helicity, uh, precip, mslp, heights_500, jet, srh). Add lat/lon + radius_km to zoom. Use cycle parameter for specific model runs.