What problem does it solve? Creating high-fidelity research and technical diagrams in draw.io is error-prone when done blind: text overflows, arrows collide with boxes, and styles drift from reference figures. This Skill turns diagram authoring into an evidence-driven loop of XML authoring, local preview, screenshot review, and defect fixing. ## Core Features & Use Cases - Editable .drawio XML authoring: Generate diagrams.net files with explicit geometry for research figures, ML architecture diagrams, and paper method overviews, with pre-flight static validation of visual quality. - Screenshot feedback iteration: Preview diagrams locally via a diagrams.net iframe, capture canvas-only screenshots, and run structured defect inventories, red-team audits, and self-scoring before handoff. - Reference replication protocol: Reproduce reference figures through required intermediate artifacts (visual-spec, layout-grid, asset-ledger, defect-log) and style extraction from provided images. - Use Case: Given a machine learning paper's method section and a style reference figure, produce an editable, top-conference-quality architecture diagram verified across multiple screenshot review cycles. ## Quick Start Use the drawio-diagram-builder skill to create an editable draw.io diagram of my paper's method pipeline and refine it with screenshot review.
