myzhao0114-del/scientific-figure-skill

scientific-figure-skill is a packaged set of rules and templates for making scientific charts that look ready for a journal paper, a thesis, or a defense slide deck. The author is a researcher who got tired of AI-generated figures that look fine at first glance but fall apart on closer inspection: mixed Chinese and English fonts, missing axis units, loud colors, sloppy p-value formatting, vague error bars, and captions that describe a plot but never state a conclusion. The project packages years of figure-fixing experience as a reusable skill. The skill is built for Codex, OpenAI's coding agent. Installation is one git clone command into the Codex skills directory. After that, the user gives the agent a short instruction in plain language, for example, use this Excel file to make a regression figure, or fit a GAM on AQI and AI and make the plot. The skill handles the rest, instead of asking the user to spell out every formatting rule. The default behavior bakes in a long list of conventions. Fonts are split into a Chinese family and an English family with cross-platform fallbacks. Colors default to restrained, low-noise palettes, with the colorblind-friendly Okabe-Ito set when more colors are needed. P-values follow standard rules such as p < 0.001 instead of p = 0.0000, and significance is marked with the usual asterisks plus ns. Error bars and captions are pushed toward one-pass correctness, and figure borders are drawn on all four sides rather than only on the left and bottom. Before plotting, the skill runs a data health check. It writes a report file called 00_data_health.md that covers missing values, outliers, collinearity by VIF, distributions, and normality. It then cleans the data, saves intermediate tables as xlsx so every number on the final chart can be traced back to a source row, draws the figure under a unified style, and exports PDF, PNG, and JPG copies along with a structured caption. The project ships a SKILL.md, references for chart style, captions, statistics, PDF layout, and xlsx export, plus an agent config for OpenAI. It is aimed at graduate students and researchers who want plotting workflows to be more standardized.