What problem does it solve? Getting research datasets into an experiment-tracking platform with proper provenance, searchability, and access control is tedious and error-prone. This Skill handles uploading dataset bytes (or registering existing gs:// or s3:// data by reference), authoring searchable metadata, and linking datasets as experiment or variation inputs in Chronicle. ## Core Features & Use Cases - Dataset Upload & Sharding: Upload a single file or a directory of shards as components via presigned PUTs, with automatic sha256 and size provenance recording. - Searchable Metadata Layer: Author a metadata document (description with LaTeX, PDE family, boundary conditions, variables, properties) that is full-text and semantically indexed for discovery. - Register by Reference: Register datasets already living in GCS or S3 without moving bytes, created ready in one call. - ACL-Aware Linking: Link datasets at experiment or variation level with control over ACL propagation for sensitive data. - Use Case: You have a 13 GB turbulence simulation in a GCS bucket. Register it by reference with rich PDE metadata, link it as a variation input, and teammates can later find it via faceted search on pde_family and precision. ## Quick Start Ask the agent to upload the dataset at ./data/fields.npz to the current Chronicle experiment with a description and metadata so it is searchable.
