Data Visualization And Figures

Purpose

Help the agent make scientific figures honest, readable, accessible, and ready for their medium by matching the visual form to the evidence and reader task.

Use this skill when

• The user asks for figure critique, chart choice, plot redesign, table redesign, graphical abstract planning, or publication-ready figure checks.
• The artifact includes distributions, p-values, effect sizes, intervals, small samples, repeated experiments, heatmaps, networks, genomic views, set intersections, or dense multi-panel displays.
• The request involves microscopy, photographs, image overlays, scale bars, insets, channels, annotations, layout, labels, typography, color, accessibility, posters, slides, or manuscript export.

Do not use this skill when

• The task is only plotting-library syntax, package errors, or data-frame debugging.
• The user wants statistical analysis design with no visual artifact or visual decision.
• The request is prose-only writing feedback with no figure, table, diagram, or visual-output concern.

Core workflow

1. Name the figure job: comparison, distribution, relationship, composition, process/overview, exact lookup, image evidence, or publication export.
2. Identify missing context that changes the recommendation: data type, n, independent unit, audience, medium, venue constraints, legend, or actual figure/image.
3. Inspect the actual figure before making exact layout, palette, microscopy, or graphical-abstract claims; if it is unavailable, state the recommendation as conditional.
4. Route to the narrowest reference file, then diagnose the highest-risk failure before polishing style.
5. Return recommendations as problem -> fix -> rationale -> priority, with assumptions and unresolved checks separated from confirmed findings.

Reference routing

• Open references/chart-selection.md for chart family, raw-pattern visibility, bar/line alternatives, tables, and ordinary plot selection.
• Open references/statistics-and-uncertainty.md for effect sizes, intervals, p-values, replicate structure, overplotting, outliers, and small-n displays.
• Open references/color-and-accessibility.md for palettes, grayscale/color-vision checks, redundant encodings, and semantic color use.
• Open references/specialized-figures.md for heatmaps, networks, genome tracks, set intersections, temporal/dense data, matrices, and 3D displays.
• Open references/biological-images.md for microscopy, image panels, scale bars, channels, insets, annotations, and image-analysis workflow reporting.
• Open references/layout-and-typography.md for multi-panel hierarchy, labels, legends, axes, callouts, salience, spacing, typography, and final-size readability.
• Open references/conceptual-figures.md for graphical abstracts, overview figures, mechanism diagrams, pathways, neural-circuit diagrams, arrows, and schematic grammar.
• Open references/publication-technical-requirements.md for manuscript, poster, slide, preprint, raster/vector, font, resolution, color-mode, and venue-specific checks.

Output formats

• Figure critique: prioritized findings in the order evidence-job mismatch, hidden data, statistical ambiguity, accessibility/color, layout/labels, export risk, and visual-review needs.
• Redesign plan: recommended form, encodings, layout changes, caveats, and required context.
• Publication-readiness pass: ready/needs revision/blocked status with must-fix items and venue assumptions.
• Before/after guidance: concise contrast between the current design and the stronger alternative.

Quick checklist

• Does the visual form match the scientific task and data structure?
• Are raw observations, sample size, spread, uncertainty, independent units, and outliers visible when they affect interpretation?
• Does color encode meaning accessibly, with redundant cues where needed?
• Are image scale, channel identity, annotations, and analysis workflow clear when images support the claim?
• Do layout, labels, axes, typography, and export settings work at final size?

Common pitfalls

• Recommending a prettier chart before naming the figure job.
• Hiding continuous or small-n data behind mean-only bars or lines.
• Treating p values, stars, or summary statistics as the visual evidence.
• Using diverging heatmap colors without a meaningful midpoint.
• Inferring image scale, channel meaning, palette accuracy, or exact layout quality without inspecting the actual figure.
• Guessing current journal requirements instead of asking for or checking the venue instructions.

Quality bar

• Give specific, executable figure advice, not taste-level comments.
• Separate confirmed visual findings from conditional guidance and missing context.
• Treat accessibility, uncertainty, replicate structure, and final-size legibility as core checks.
• Keep provenance, source names, and extraction notes out of user-facing responses.