Changes in version 0.1.0                        

Bug fixes

  - on_files() (and therefore on_download(), which lists files before
    fetching) no longer fails with HTTP 400 GRAPHQL_VALIDATION_FAILED.
    The getFiles query requested a key field that the current OpenNeuro
    schema no longer exposes on DatasetFile; the query now requests id
    (the directory tree token) and urls (direct HTTPS download links)
    instead (#1).
  - on_files() now returns an id column (the token to pass back as tree
    when recursing into a directory) and a urls list column. The key
    column is retained as a backward-compatible alias of id.
  - GraphQL validation/HTTP errors are no longer reported as "Network
    error connecting to OpenNeuro / Check your internet connection." The
    API's actual error message (e.g. a GRAPHQL_VALIDATION_FAILED reason)
    is now surfaced, and genuine connectivity failures remain distinct
    (#1).
  - on_doctor() and backend auto-selection no longer treat a broken AWS
    CLI as available. The S3 backend check now runs aws --version and
    requires a clean exit, instead of only looking for the aws binary on
    PATH (#1).

Initial Release

  - Search and explore OpenNeuro datasets via GraphQL API
    
      - on_search() - Full-text and modality-based search
      - on_dataset() - Detailed metadata retrieval
      - on_snapshots() - List version history
      - on_files() - Browse file trees

  - Download datasets with multi-backend support
    
      - on_download() - Download files with automatic backend selection
      - HTTPS fallback always available
      - S3 (via AWS CLI) for faster parallel downloads
      - DataLad for version-controlled provenance tracking

  - CRAN-compliant caching with manifest tracking
    
      - on_cache_list(), on_cache_info(), on_cache_clear()
      - Platform-appropriate cache locations via tools::R_user_dir()
      - Atomic writes prevent corrupt manifests
      - Resume support for large downloads

  - Lazy handle pattern for pipeline integration
    
      - on_handle() - Create lazy dataset references
      - on_fetch() - Materialize data when needed
      - on_path() - Get local paths for immediate use

  - Developer utilities
    
      - on_client() - Create and configure API clients
      - on_doctor() - Diagnose backend availability
      - Comprehensive mocked test suite with httptest2