jx: an agent-composable marimo notebook catalog for scientific data exploration

jx is an experiment in agent-driven scientific data exploration. It pairs a catalog of runnable marimo notebooks — each embodying a real analysis use case and exposing reusable pure functions — with a thin skill file that lets an AI agent compose new analyses from existing notebooks in response to a user's question. The catalog is simultaneously documentation, working code, and a compositional substrate for agents, without the maintenance overhead of a separate library. The initial instantiation targets a large public morphological profiling dataset, but the pattern is dataset-agnostic.