| name | academic-figure-workflow |
| description | Plan, create, revise, and validate academic figures for papers, theses, dissertations, and technical reports. Use for SVG mechanism diagrams, conceptual models, Nature-grade multi-panel Matplotlib figures, data-driven plots,正文 figure interpretation handoff, draw.io engineering diagrams, workflow charts, use-case diagrams, architecture diagrams, tables-as-figures, figure captions, publication visual standards, export/editability QA, accessibility/color checks, and user-confirmed OpenAI image generation. Chinese triggers: 论文作图, 机理图, SVG绘图, Nature风格多面板图, matplotlib作图, 数据图正文分析, 图的作用阐释, drawio流程图, 用例图, 架构图, 论文插图标准, 图片投稿标准, 大模型绘图API. |
Academic Figure Workflow
Use this skill for academic and thesis figures. Choose the method by figure type, evidence needs, editability, and reproducibility. For data-backed plots, also prepare the paragraph-level figure argument so正文 text can explain the plot's purpose, quantitative pattern, and implication.
Boundaries
Do:
- Plan figure claims, panels, captions, source files, and exports.
- Create SVG mechanism figures, Draw.io engineering diagrams, and reproducible Matplotlib figures.
- Validate that every visual element traces to evidence, data, code, manuscript text, or user-approved facts.
- For data-backed plots, act as a visualization advisor first: inspect the data and figure argument before choosing or drawing a chart.
Do not:
- Invent data, mechanisms, system modules, experimental results, labels, or visual evidence.
- Use AI image generation without the explicit gate in
ai-image-generation-gate.md.
- Call a figure submission-ready when labels overlap, source files are missing, or evidence trace is incomplete.
Intake
Collect:
- figure purpose and target chapter or section;
- audience and manuscript language;
- source facts: text, evidence table, codebase facts, data columns, equations, or rough sketch;
- target examples, journal/school guide, or lab style guide when available;
- target venue stage: draft, thesis insertion, initial submission, final accepted artwork, or repository supplement;
- output format: SVG, Draw.io, PNG, PDF, DOCX insertion, or source plus export;
- style constraints: grayscale, journal style, school template, colour restrictions, font, size;
- AI image generation permission when relevant.
Before drawing, plotting, generating, or exporting, ask for missing materials that would materially improve figure quality:
Required or high-value materials:
Missing materials:
Can proceed without them: yes/no
Fallback if unavailable:
If key evidence is missing, stop at a figure specification, storyboard, or provisional template until the user confirms execution.
Method Selection
| Figure type | Default method |
|---|
| Mechanism diagrams, conceptual models, moderately complex scholarly schematics | SVG |
| Workflow, use-case, module, architecture, ER/data-flow, software process diagrams | Draw.io |
| Data-driven plots or multi-panel quantitative figures | Data-visualization advisor flow plus Matplotlib Nature-grade protocol with source data, editable SVG/PDF export, and final raster preview |
| Photorealistic, illustrative, texture-like, or hard-to-specify visual assets | OpenAI image generation after user confirmation |
| Simple numeric comparison | Native table or chart in manuscript tool |
Read figure-method-selection.md when a figure could be created in more than one way.
Core Rules
- Draw only what the evidence, codebase, experiment, or user-approved concept supports.
- Make every panel serve a scientific claim, method step, result contrast, or design decision.
- Prefer editable source files:
.svg for SVG and .drawio for Draw.io.
- Export publication formats only after the editable source is accepted.
- Keep labels concise and consistent with manuscript terminology.
- Record assumptions and unresolved evidence instead of filling gaps visually.
- Ensure figure text remains readable at final DOCX/PDF insertion size and remove text that should not appear inside the figure.
- Prefer Draw.io for workflow and process diagrams; preserve user-edited
.drawio styling unless a redraw is explicitly requested.
- Preserve formulas, superscripts/subscripts, and notation in labels, captions, tables-as-figures, and exports.
Workflow
- Define the figure claim, target section, source facts, method, output files, and validation checks.
- For complex, multi-panel, submission-bound, code-backed, or data-backed figures, apply figure-argument-contract.md.
- For submission-quality or journal-facing figures, apply publication-visual-standards.md, export-and-editability-standards.md, and accessibility-and-color-standards.md.
- For data-backed plots, chart-choice questions, user-provided CSV/Excel/DataFrame files, or risky user-specified chart types, apply data-visualization-advisor.md before plotting.
- For正文 narration of a data-backed plot, also apply the writing-workflow reference for data-figure body analysis so the manuscript states the figure role, best/worst region, concentration or sparsity, and supported conclusion.
- For Nature-style or multi-panel Matplotlib figures, apply nature-matplotlib-protocol.md, multi-panel-layout-patterns.md, and chart-type-visual-standards.md.
- For SVG mechanism diagrams or Draw.io engineering diagrams, apply svg-and-drawio-diagrams.md; for Draw.io code-backed or engineering diagrams, also apply drawio-engineering-diagram-standards.md.
- For image plates, photographs, microscopy, heatmaps over images, or mixed raster/vector panels, apply image-panel-integrity.md.
- For AI-generated bitmap assets, apply ai-image-generation-gate.md before generation.
- When a venue-specific requirement, external visual example, or numeric artwork standard matters, apply publisher-visual-source-map.md and verify the current official guide when possible.
- Apply notation-and-conversion-integrity-gate.md when figure labels, formulas, superscripts/subscripts, citations, or cross-references may be converted.
- For any submission-quality generated or revised figure, apply rendered-figure-qa.md before calling it ready.
- Provide the editable source, exports if generated, caption draft, placement recommendation, evidence/code trace, rendered QA result, and unresolved issues.
Validation
Before finishing:
- confirm the method matches the figure type;
- inspect generated source files when possible;
- check label overlap, readability, export integrity, and editability;
- inspect a rendered PNG preview at intended final size for submission-quality figures;
- verify final-size readability, font consistency, line weight, colour meaning, grayscale or colour-blind-safe interpretation, and caption/description handoff when relevant;
- report the editable source path, especially
.drawio paths for workflow diagrams;
- trace each panel, node, edge, label, plotted value, or annotation back to approved source facts;
- confirm code-backed diagrams match actual implementation;
- defer figure numbering or final caption style to formatting when the target format is not known;
- list unresolved scientific, data, style, or export issues separately.
Reference
Read figure-method-selection.md for method-choice rules.
Read figure-argument-contract.md for complex, evidence-backed, multi-panel, or submission-bound figures.
Read publication-visual-standards.md for journal-facing typography, sizing, line weight, spacing, and visual hierarchy.
Read export-and-editability-standards.md for SVG/PDF/EPS/TIFF/PNG choice, editable text, embedded fonts, RGB, DPI, and file packaging.
Read accessibility-and-color-standards.md for colour-blind-safe design, grayscale checks, direct labels, contrast, and figure descriptions or alt text.
Read data-visualization-advisor.md before choosing or drawing data-backed plots from CSV/Excel/DataFrame files, when the user asks what chart to use, or when a requested chart type may be misleading.
Read nature-matplotlib-protocol.md for Nature-grade Matplotlib figures.
Read multi-panel-layout-patterns.md for hero panels, asymmetric layouts, shared legends, panel order, and anti-redundancy layout rules.
Read chart-type-visual-standards.md for bars, lines, scatter, heatmaps, matrices, ablations, comparisons, and tables-as-figures.
Read svg-and-drawio-diagrams.md for mechanism diagrams, conceptual SVGs, architecture diagrams, and code-backed Draw.io diagrams.
Read drawio-engineering-diagram-standards.md for Draw.io architecture, workflow, module, use-case, data-flow, and code-backed diagram conventions.
Read notation-and-conversion-integrity-gate.md when figure text, notation, formulas, superscripts/subscripts, citations, or cross-references are converted across figure, manuscript, DOCX, PDF, SVG, or Draw.io outputs.
Read image-panel-integrity.md for photographs, microscopy, image plates, raster panels, scale bars, crop/manipulation notes, and mixed raster/vector figures.
Read ai-image-generation-gate.md before using AI image generation.
Read publisher-visual-source-map.md when mapping standards to a specific publisher, journal family, venue, or external figure-standard source.
Read rendered-figure-qa.md after generating or revising a submission-quality figure to run the rendered-preview QA loop before final export or delivery.
Bundled Utilities
scripts/make_nature_multipanel.py: generate Nature-style multi-panel Matplotlib figures from a panel spec and source data; exports editable SVG/PDF, PNG preview, and QA reports.
scripts/nature_mpl_style.py: shared Nature-style Matplotlib defaults, semantic palette, panel labels, sizing helpers, and export helpers.