earth2studio-create-datasource

Follow the guided workflow: analyze the remote store, choose the source type (DataSource, ForecastSource, DataFrameSource, or ForecastFrameSource), create a lexicon mapping variables, implement the async source class, register it, and add tests. The skill provides skeleton templates for each step.

DataSource returns an xr.DataArray for gridded analysis or reanalysis data without lead_time, while ForecastSource includes a lead_time dimension for gridded forecast data. Sparse observation sources use DataFrameSource or ForecastFrameSource returning pd.DataFrame instead.

Prefer fsspec-based filesystems: s3fs for AWS S3, gcsfs for Google Cloud Storage, adlfs for Azure, and fsspec's HTTPFileSystem for HTTP. Avoid boto3, google-cloud-storage, or raw requests unless fsspec cannot access the store.

Required tests include a slow network fetch test, a cache toggle test, a mandatory mock test with no network access, exception handling tests, and an available() classmethod test. Target at least 90% line coverage when running with the --slow flag.

Do not use it for fetching data with existing sources, running model inference, or installation tasks. It handles one source type per invocation and requires network access for the real-data validation step.