paper-figure

The paper-figure skill is a specialized agentic tool designed to bridge the gap between raw experiment data and submission-ready academic visualization. It is intended for researchers and ML engineers who need to convert JSON, CSV, or log-based experiment results into professional, consistent graphical representations. By automating the boilerplate of Matplotlib and LaTeX formatting, it allows users to focus on scientific narrative rather than low-level styling code.

• Automatically generates publication-quality line plots, bar charts, scatter plots, heatmaps, and box/violin plots with pre-configured publication-style aesthetics.

• Creates professional-grade LaTeX tables for ablation studies, method comparisons, and prior bounds evaluation.

• Supports multi-panel figure orchestration, enabling the aggregation of multiple plots into consistent grids suitable for paper layouts.

• Implements strict visual consistency via a shared paper_plot_style.py, including font, DPI, color palettes, and axis management compliant with top-tier conference standards.

• Integrates with research planning workflows by parsing PAPER_PLAN.md to prioritize figures based on high-priority research findings.

• Utilizes external reviewer models (via Codex MCP) to verify figure clarity and data representation accuracy before final export.

• Users should provide input data such as JSON results, CSV files, or experiment logs in the project root or figures/ directory.

• The skill distinguishes between data-driven plots and manually created figures; it preserves existing content in the figures/ directory, meaning users should place architecture diagrams, screenshots, or hero figures there manually.

• The output is highly customizable via constants, allowing users to toggle between publication, poster, and slide visual presets.

• While it can generate TikZ skeletons for pipeline diagrams, complex architecture drawings remain a manual process utilizing tools like draw.io or Figma.

• This skill is best used in tandem with other research agent workflows like paper-plan or paper-write to maintain a seamless, automated end-to-end research loop.