What problem does it solve? Research figures often lack traceable provenance, cannot be regenerated from scripts, or carry captions that overstate the underlying statistics. This Skill governs figure classification, provenance binding, and caption-claim checks so that paper figures are reproducible and claims stay proportional to the evidence. ## Core Features & Use Cases - Figure Classification: Assigns every figure to a class (paper_figure, diagnostic_figure, exploratory_figure, pipeline_diagram, reviewer_response_figure, website_figure) with class-specific governance rules. - Provenance & Regeneration Binding: Requires paper figures to be bound to a script path, input artifact, and reproducible command, recording parameters and seeds. - Caption-Claim Proportionality: Checks that captions do not claim more than the underlying statistic supports, and propagates PROVISIONAL flags from input artifacts to figures. - Use Case: A striking exploratory UMAP plot is proposed as a paper figure; the Skill blocks direct promotion and requires a validation check, a regeneration script, an input artifact, and a proportional caption first. ## Quick Start Ask the assistant to classify a proposed paper figure and verify that it can be regenerated from a script with a proportional caption.
