scipilot-figure-skill

Why we included this project

Researchers who already run their work inside Claude Code, Codex, or Cursor get a real shortcut at the point where most papers stall: turning analyzed data into a figure that journals will accept. The skill profiles the data first, checking column types, sample sizes, distributions and outliers, then asks what point the figure is supposed to make before recommending a chart type, so it tends to catch choices that generic plotting tools never question, like mean bars drawn from five samples. When it finally renders, it applies preset styles for Nature, IEEE, Elsevier, PNAS and Chinese journals, sets up CJK fonts so Chinese labels come out as characters instead of empty boxes, and runs a visual audit that flags clipped text, overlapping legends and misaligned panel labels before you export. Because it installs as a skill folder rather than a separate pipeline, it drops into an existing assistant setup and earns its keep once you are producing many figures that need to look consistent.