From Paper Draft to Publication Figures: academic-plotting

Where this fits

You just finished a NeurIPS draft and need to replace a blurry pipeline diagram plus regenerate an ablation plot from experiment CSVs. Instead of toggling between diagram tools, matplotlib scripts, and LaTeX, you hand the paper section and results to an agent powered by academic-plotting. It extracts system components and relationships for Gemini-based architecture diagrams, then auto-selects the right chart type for data figures using matplotlib/seaborn.

Why agents benefit

• Can ingest an arbitrary methods section and produce an architecture diagram without requiring the user to manually specify every box, arrow, and connection.
• Auto-selects chart type from the data shape: line charts for training curves, grouped bars for ablations, heatmaps for matrices, and scatter plots for parameter analyses.
• Uses pinned plotting dependencies aligned with paper-style output, making it easier to tune font sizes, image dimensions, and LaTeX compatibility in one workflow.
• Separates generative diagram creation from deterministic data plotting, giving the agent a clear decision boundary: Gemini for spatial layouts, matplotlib/seaborn for quantitative figures.
• Can be invoked at any writing stage, from initial experiment dump to final camera-ready revision, without manual export/import loops.

Practical scenarios

Rebuilding an architecture diagram from a paper draft

An agent reads a paragraph describing a transformer encoder with cross-attention and a classification head, then generates a diagram with labeled components and arrows for a quick visual check before polish.

Turning an ablation table into a publication chart

Given CSV rows of per-module ablations, the agent selects a grouped bar chart or heatmap, labels the axes, and exports a high-resolution figure with consistent styling for the appendix.

Reproducing figures for a new dataset

The agent reruns experiments on a new dataset and reuses the same skill to regenerate comparison curves, ensuring the main paper and supplementary material keep the same visual style.

Add it to your agent workflow

Install the skill with:

npx skills add OpenRaiser/NanoResearch --skill academic-plotting

Then point the agent to a paper section or an experiment CSV. The skill decides which workflow to use: Gemini for diagrams, matplotlib/seaborn for data figures. Keep generated scripts in a figures/ directory so the next revision can be updated without starting over.

Compare before adopting

Before adopting, compare quality signals: does the output look like a real NeurIPS figure or only a rough schematic? Check maintenance freshness by looking at recent commits and dependency pins for google-genai, matplotlib, and seaborn. Consider alternatives such as direct Gemini calls for diagrams or your own plotting templates. Also assess workflow fit: if your lab enforces strict style conventions, you may need to add a post-processing step around the skill's default output.

Why it is worth tracking

This skill targets a pain point generic coding agents often miss: producing publication-specific visual output rather than generic charts. With 1,362 stars on GitHub and an active repository, it shows clear community momentum. Check the skill's page at https://www.openagentskill.com/skills/openraiser-academic-plotting to compare versions and community adoption before wiring it into a recurring research pipeline. Evaluate it when your agent needs consistent figures across experiments, and track updates for changes to model defaults and plotting behavior.