Imported from tiangong-ai/agent-skills (open-meteo-historical-fetch/SKILL.md). Install upstream with npx skills add tiangong-ai/agent-skills --skill open-meteo-historical-fetch. Copyright stays with the author.

Open-Meteo Historical Weather

Use the CLI-owned open-meteo.historical-weather capability. This Skill supplies intent routing and result-use boundaries only; the CLI owns source discovery, input/output schemas, HTTP behavior, limits, validation, and receipts.

Before running

1. Read references/tiangong-data-requirement.json.
2. Use the caller- or workspace-resolved stable CLI. The requirement declares compatible capability and operation contract majors; it does not select a package build.
3. Run data describe with that same CLI. Continue only when the capability ID, required contract majors, and the open-meteo.series-all-null operation feature match; then copy the exact current capability/operation versions from that response into the run request.
4. Run the default static doctor. Do not add --live unless the user explicitly asks for a provider probe.

tiangong-ai data describe open-meteo.historical-weather --json
tiangong-ai data doctor open-meteo.historical-weather --json

Use the returned Discovery Metadata to confirm current model coverage, resolution, update delay, variable availability, public-endpoint terms, attribution, provides, and doesNotProvide. Do not substitute facts remembered from an older Skill revision.

Prepare the request

Build a tiangong.data.run-request.v1 envelope and replace the two version placeholders with the exact versions from the same data describe response. Coordinates are WGS84 decimal degrees. Both variable arrays are required; use an empty array for an unneeded granularity. At least one array must be non-empty. The CLI fixes timezone and units, preserves coordinate order, and normalizes variable order:

{
  "schemaVersion": "tiangong.data.run-request.v1",
  "capabilityId": "open-meteo.historical-weather",
  "capabilityVersion": "<describe.manifest.capabilityVersion>",
  "operationId": "fetch",
  "operationVersion": "<describe.manifest.operations[0].operationVersion>",
  "input": {
    "locations": [
      { "latitude": 52.52, "longitude": 13.41 },
      { "latitude": 48.85, "longitude": 2.35 }
    ],
    "startDate": "2024-01-01",
    "endDate": "2024-01-07",
    "hourlyVariables": [
      "temperature_2m",
      "relative_humidity_2m",
      "precipitation",
      "soil_moisture_0_to_7cm"
    ],
    "dailyVariables": [
      "temperature_2m_max",
      "temperature_2m_min",
      "precipitation_sum",
      "et0_fao_evapotranspiration"
    ],
    "model": "era5",
    "cellSelection": "land"
  }
}

Use the operation input schema returned by the same data describe response for current variable codes, model meanings, and limits. Select era5 or era5_land for a multi-decade analysis where one consistent model family matters. The default best_match favors available local detail but can change model families over time. Do not add an API key: this capability uses the public non-commercial endpoint only. Commercial customer-endpoint access requires a separately reviewed capability. Do not silently geocode names, adjust coordinates, widen the date range, switch models, or add variables beyond the user's intent.

Run

tiangong-ai data run open-meteo.historical-weather fetch \
  --input /absolute/path/to/request.json --json

The command emits a tiangong.data.run-result.v1 envelope. Preserve its contract, warnings, errors, and receipt with data when handing the result to another workflow.

Result boundaries

• Treat values as gap-filled reanalysis or model-grid estimates, not raw observations from a named station. Preserve requested and returned grid coordinates plus elevation.
• Retain nulls as unavailable model values. Never convert them to zero or interpolate them without an explicit downstream method.
• Distinguish series-missing from series-all-null: missing means the provider did not return a requested series, while all-null means it returned a structurally valid, aligned series with no usable numeric values. Preserve an all-null series and report the machine-readable issue code.
• Treat partial as incomplete coordinate, section, timestamp, variable, or unit coverage and report the affected paths with the usable series.
• In GMT mode, require exactly 24 strictly ascending hourly timestamps and one strictly ascending daily date per inclusive request date, with a zero provider UTC offset. Preserve count, ordering, timezone, unit, alignment, and non-numeric-value issues; a short axis is not complete coverage.
• Treat blocked as no usable business result and surface the structured errors instead of bypassing limits or switching endpoints.
• Report record-limit truncation. The CLI keeps every retained variable aligned to its returned time axis, but a truncated result is not exhaustive.
• Preserve the selected model in analysis and citations. Do not combine models or interpret a Best Match discontinuity as a weather or climate trend.
• Choose a station source when instrument provenance, local measurement quality, or regulatory-grade observations are required.
• Do not use this capability for forecasts, future climate scenarios, causal attribution, significance testing, or safety-critical weather decisions.
• The fixed public contract intentionally omits the source script's multi-model request, arbitrary timezone and units, optional API-key/customer access, endpoint overrides, and raw JSON/log artifacts. Expand those only through a separately reviewed CLI contract.
• Attribute Open-Meteo and the underlying data providers. The public endpoint is non-commercial; do not imply commercial-use permission.
• Cross-source comparison, statistical inference, and research evidence admission belong to the caller or Auto Research, not this atomic Skill.

Reference

• references/tiangong-data-requirement.json: stable capability requirement; it is not a package lock.