What problem does it solve? Academic papers need publication-quality figures, but AI-generated diagrams often come out as flat box-and-arrow flowcharts with garbled labels, fabricated content, or non-editable raster pixels. This Skill fills empty LaTeX figure placeholders with real, grounded, editable vector figures. ## Core Features & Use Cases - Grounded image-model rendering: Designs a concrete visual blueprint, grounds every schematic on a real top-venue paper's main figure fetched via WebSearch and fetch_reference_figures.py, then renders through the official PaperBanana pipeline or the built-in gen_image.py image-model client. - Enforced multi-round vision critique: Claude inspects each rendered PNG with its own vision over at least 2 logged rounds, checking faithfulness, readability, richness, and integrity before approval. - Editable vector output: Approved PNGs are redrawn as native SVGs by the sibling ts-figure-svg skill and converted to vector PDFs; matplotlib is used only for real measured-data results plots. - Use Case: While compiling a journal paper, each \fbox placeholder in sections/*.tex is classified by section, rendered with the appropriate engine, critiqued, vectorized, and inserted as an extension-less \includegraphics with a manifest entry. ## Quick Start Fill every empty figure placeholder in my paper's LaTeX sections with grounded, publication-quality figures and insert them as editable vector PDFs.
