What problem does it solve? Creating publication-quality scientific figures (flowcharts, neural network architectures, biological pathways, circuit diagrams) normally requires manual drawing tools and design expertise. This Skill generates them from natural-language descriptions and automatically reviews each image against a quality threshold for your document type. ## Core Features & Use Cases - Natural-language diagram generation: Describe a CONSORT flowchart, Transformer architecture, MAPK pathway, or system block diagram and receive a PNG via the Nano Banana 2 image model on OpenRouter. - Smart iterative refinement: Gemini 3.6 Flash scores each image against document-type thresholds (journal 8.5/10 down to presentation 6.5/10) and regenerates only when below threshold, capped at 2 iterations. - Auditable review logs: Every run writes versioned images plus a JSON review log with scores, critiques, and early-stop reasons; failed reviews are recorded as unmeasured rather than given a fabricated score. - Use Case: A researcher needs a CONSORT participant flow diagram for a journal submission. They describe the screening, exclusion, and randomization counts, run the script with --doc-type journal, and get a reviewed PNG plus an auditable score log. ## Quick Start Ask the agent to generate a scientific diagram, for example: create a CONSORT participant flow diagram with 500 screened, 150 excluded, and 350 randomized, saved to figures/consort.png for a journal paper.
