How to use the Scientific Figure Design agent skill

What the skill makes the agent do

When the skill is installed and you ask the agent to plot research data for a paper, it follows a set of publication-figure guidelines. It picks journal-appropriate figure sizes, uses colorblind-friendly palettes, adds error bars and significance markers, labels panels (A), (B), (C), and checks that axis labels include units.

The skill is only a set of written best practices. It does not add new software tools of its own.

• Figure sizes: single column about 3.3-3.5 inches (85-89 mm), 1.5 column 4.5-5.5 inches (114-140 mm), double column 6.5-7.1 inches (165-180 mm).
• Resolution: 300 DPI minimum for raster images; vector formats (PDF, EPS, SVG) are preferred.
• Colour: seaborn "colorblind", Okabe-Ito, viridis or cividis palettes, and no reliance on colour alone.
• Statistics: error bars with the type stated in the caption (SEM, SD, 95% CI), and significance stars (* p<.05, ** p<.01, *** p<.001).
• Export checks: embedded fonts or outlined text, units in axis labels such as "Time (s)", and captions that explain every symbol and panel.

Install the skill

You can install the skill in one of three ways. Download the ZIP from this page and unzip it into your personal skills folder ~/.claude/skills/, or into a project's .claude/skills/ folder if you want it only for one project. Alternatively, upload the skill in the Claude apps under Settings → Capabilities → Skills.

The skill's files are SKILL.md. Once it is in place, the agent can load it when your request matches its description.

• Personal install: unzip into ~/.claude/skills/.
• Project install: unzip into the project's .claude/skills/ folder.
• Claude apps: Settings → Capabilities → Skills, then upload.

Trigger the skill with natural requests

You do not need a special command. The skill's description covers creating journal submission figures with proper formatting, accessibility and statistical annotations, and its trigger keywords include figure, plot, chart, visualization, matplotlib, seaborn, colorblind and publication. Name the paper context and the data in your request so the agent knows the figure is for publication.

Requests that work well describe the figure, the data and the venue, for example asking for a two-group comparison plot for a manuscript with error bars and significance marks. Mentioning the target journal helps, but the skill says to check that journal's author guidelines, which take priority.

• "Plot these reaction times as a box plot with individual points for my paper, with error bars and significance stars."
• "Make a three-panel figure for a journal submission, labelled A, B, C, with colourblind-safe colours."
• "Create a matplotlib figure at single-column width for my thesis chapter and check the axis units."

Examples of what the output looks like

For a small-sample comparison, the guidelines favour a box plot or dot plot with the individual points visible rather than a bar chart. For a multi-panel figure, the skill recommends plt.subplots() with constrained_layout=True, consistent axis scales when comparing related data, and bold uppercase panel labels in the top-left corner.

Before you submit, ask the agent to run through the export checklist: vector format for line art, fonts embedded or converted to outlines, units in axis labels, and captions that match the colours and symbols in the figure.

What you need

The skill needs Python with matplotlib and seaborn, but only if the agent writes and runs plotting code. The skill itself runs no scripts and makes no network calls. Check that Python and both libraries are available in the environment where the agent works.

• Python with matplotlib and seaborn (needed only for code the agent writes and runs).
• No scripts are run by the skill.
• No network access is used by the skill.

Safety notes and limits

The skill runs no scripts and uses no network, so it does not send your data anywhere on its own. Any code the agent writes to make a plot runs in your environment, so review it as you would any code before running it on sensitive data.

The rules are tuned for academic journals. For business dashboards or marketing graphics, some advice, such as removing gridlines and using direct labels, may not suit your purpose. The guidance says each target journal's author guidelines override these defaults.

• Always check the target journal's author guidelines before final submission.
• Do not use this skill as the final word on formatting for business or marketing graphics.

Tips for better results

Give the agent exact data details: sample size, what the error bars represent, and which comparisons are significant. The skill asks for error type in captions and avoids bar charts for small-n data, so these details shape the output.

Test figures for colourblind readability and check them in grayscale for print. The skill recommends a colourblind simulator before submission, which the agent cannot fully replace.

• State the sample size and the statistic each error bar shows.
• Ask for the figure at a specific column width.
• Ask for a final checklist pass before you export.