What problem does it solve? Image models produce raster PNGs of paper figures that are not editable, often contain fabricated labels or wrong topology, and cannot be fixed or restyled. This Skill turns a generated figure candidate into a native SVG where every label is live text and every element traces back to the paper's actual method, then verifies it on the compiled PDF page. ## Core Features & Use Cases - Two-stage figure pipeline: Runs the official PaperBanana engine (Retriever→Planner→Stylist→Visualizer→Critic) to generate candidate PNGs, then extracts the design language into a style sheet and redraws the figure from the paper's facts as real SVG elements. - Measured repair loop: audit_svg.py checks structure and geometry without a renderer — catching overflow, text overlap, stroke-scaled arrowheads, dangling connectors, font fallback risks, and traced path-soup — across at least four rounds with per-round snapshots. - Page-level acceptance: Converts SVG to vector PDF with font embedding verified via pdffonts and pdfimages, measures the smallest rendered word in points, and validates a fresh rebuild against the baseline. - Use Case: You have a LaTeX paper with a placeholder figure. Use this Skill to generate candidates from your method section, pick one, and receive an editable SVG plus embedded vector PDF that passes machine-checked integrity gates. ## Quick Start Set up the PaperBanana engine with your OpenRouter API key, then ask the assistant to turn the placeholder figure in your paper's LaTeX source into an editable SVG figure.
