from claude-code-templates25,397
Generate publication-quality scientific diagrams via AI with iterative quality review and smart regeneration for journal, conference, poster, and presentation o
Generates publication-ready scientific schematics (flowcharts, neural network architectures, biological pathways, circuits, system diagrams) using an AI generation + review loop. The skill leverages Nano Banana Pro for image generation and Gemini 3 Pro for quality scoring, applying smart iteration so images are only regenerated when quality falls below document-type thresholds. Outputs include versioned images and detailed review logs.
Use this skill when you need high-quality figures for papers, posters, presentations, or reports—especially when time-saving automated layout, labeling, and accessibility (colorblind-safe palettes, typography) matter. Good for researchers preparing manuscripts, presenters building figures, and developers automating diagram creation.
Works with agent environments that can call external AI image generation services and shell scripts (Claude Code, CLI-driven agent runtimes, Python-capable agents).
Skill generates scientific diagrams via OpenRouter API (Nano Banana Pro for image gen, Gemini 3 Pro for quality review) with smart iterative refinement. Scripts are well-structured Python with argparse, type hints, and a clean class-based design. Both scripts require mandatory args (prompt, -o) and failed without them as expected — no crashes. Requires OPENROUTER_API_KEY and paid API calls to function. SKILL.md is overly long and monolithic, mixing examples, troubleshooting, and checklists that should be in references/.
No security concerns beyond normal API usage. The HTTP-Referer header leaks project info (github.com/scientific-writer). Scripts use subprocess.run with list args (safe, no shell injection). API key handled via env var/CLI flag, never hardcoded. The skill is functional but niche and requires paid external services.
