Imported from cvcyz/evidence-governed-paper-skills (
skills/paper-figure-studio/SKILL.md). Install upstream withnpx skills add cvcyz/evidence-governed-paper-skills --skill paper-figure-studio. Copyright stays with the author.
Paper Figure Studio
Do not start rendering a figure until the author approves a figure decision card.
Decision card
Ask or infer from approved paper design:
- What single reader conclusion should this figure support?
- Which evidence and variables are eligible?
- Which visual forms fit the data: comparison, distribution, relationship, time evolution, spatial structure, hierarchy, or process?
- Offer a small set of suitable figure forms and explain their trade-offs.
- Ask for the production backend: Draw.io, Origin, Python, or a combined workflow.
- State panel structure, axes/units, groups, uncertainty treatment, annotation rules, colors, target width, and file deliverables.
Record this in assets/figure-decision-card.md; wait for the author's approval before generating files.
Backend routing
- Draw.io: use for flowcharts, technical routes, architecture, editable redraws, and diagrams. For a raster reference, first inventory its elements and layout, then reconstruct it as editable objects.
- Origin: use when the author needs a directly editable Origin project. Invoke
origin-paper-figure-backendafter approval. - Python: use for transparent, reproducible data-driven plots; keep code, source mapping, and exports together.
Universal visual checks
- Size artwork for its final manuscript column width. At final placement, labels should normally read near body-text size, often about 8–12 pt depending on journal layout.
- Use English labels, units, readable line widths, color-blind-safe contrast, compact whitespace, and no collisions between text, data, legends, and axes.
- Export a vector PDF/SVG where appropriate and a 300 dpi PNG fallback; inspect the figure at final manuscript size, not only at screen size.
- Verify that visual wording and data match the approved claim and figure role.