scientific-figure-making
Skill4 starsChanged 41 days agoMITView on GitHub
Create publication figures from data with claim-aware charts, honest scales and uncertainty, export, and visual QA. Use for plots, not diagrams.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- Grady10086/Prismaticatheir own version, first seen Nov 15, 2026
- Gavin-Tao/claude-skillstheir own version, first seen Nov 2, 2026
- guoyingwei6/scientific-research-skillstheir own version, first seen Jun 16, 2026
- ChenLiu-1996/figures4paperstheir own version, first seen Apr 19, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- ChenLiu-1996/figures4papers★ 12ksame name
- Gavin-Tao/claude-skills★ 3same name
- Grady10086/Prismatica★ 2same name
- guoyingwei6/scientific-research-skills★ 2same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository19 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- academic-humanizerSkill
- citation-verificationSkill
- claim-numerics-auditSkill
- cover-letterSkill
- experiment-planSkill
- experiment-runnerSkill
- journal-scoutSkill
- paper-auditSkill
- paper-illustrationSkill
- paper-planSkill
- paper-writeSkill
- research-data-auditSkill
- research-ideationSkill
- research-publicationSkill
- research-quality-gateSkill
- research-refineSkill
- result-to-claimSkill
The file
We split this skill into the rules it gives an agent
7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 12
Use for reproducible quantitative figures such as lines, scatter plots, grouped comparisons, distributions, box/violin/point plots, …
- Distinctive: line 14
Do not use for:
- Very specific: line 29
Use scripts/profile_data.py or an equivalent reproducible inspection to check variable types, units, missingness, sample size, grouping, …
- Very specific: line 33
Read references/chart_selection.md and references/viz_pitfalls.md when the chart is not already fixed by an approved plan.
- Distinctive: line 39
Check the target venue's current dimensions, font, line-art, color, and export requirements.
- Distinctive: line 41
Do not impose universal colors for “proposed,” “baseline,” or “improvement.” Semantic mappings must fit the figure and remain consistent …
- Very specific: line 47
Use references/plot_recipes.md only for the selected chart.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Scientific Figure Making
Choose what the figure must establish before drawing it. A polished plot that answers the wrong question or hides uncertainty is a failed figure.
Scope
Use for reproducible quantitative figures such as lines, scatter plots, grouped comparisons, distributions, box/violin/point plots, heatmaps, error bars, and multi-panel layouts.
Do not use for:
- architecture, flow, concept, or process diagrams — route to the relevant diagram/illustration capability;
- interactive dashboards;
- 3D/GIS work;
- bitmap illustration or decorative scientific imagery.
Workflow
1. Freeze purpose and provenance
State the reader question, Claim/Chain ID, takeaway the evidence may support, data source, analysis unit, and intended paper location. A takeaway is provisional until the rendered figure and source data agree.
2. Profile the data
Use scripts/profile_data.py or an equivalent reproducible inspection to check variable types, units, missingness, sample size, grouping, distribution, outliers, dependence, and uncertainty inputs. Do not silently drop observations or change an analysis unit for visual convenience.
3. Select the chart
Read references/chart_selection.md and references/viz_pitfalls.md when the chart is not already fixed by an approved plan. Choose the chart from the variable relationship and claim, not from a preferred house style.
When sample size is small or distribution shape matters, show observations or distributions instead of hiding them behind a mean bar. Avoid dual axes, rainbow scales, categorical points joined as continuous trends, and other encodings that invite a false comparison unless a documented reason and clear safeguards exist.
4. Apply venue and project style
Check the target venue's current dimensions, font, line-art, color, and export requirements. Preserve an established accessible project style when present. Otherwise choose a restrained, colorblind-aware palette with redundant encodings such as markers, line styles, hatching, or direct labels.
Do not impose universal colors for “proposed,” “baseline,” or “improvement.” Semantic mappings must fit the figure and remain consistent within the manuscript.
scripts/setup_style.py provides configurable defaults and CJK fallback handling; inspect the rendered fonts and override defaults when venue or project requirements differ.
5. Draw from recorded data
Use references/plot_recipes.md only for the selected chart. Keep plotting code beside the exported figure or in the project's documented figure source location. Captions state the uncertainty definition, sample size or analysis unit when material, and any transformations or exclusions.
Scales must support an honest comparison:
- bar lengths normally require a meaningful zero baseline;
- line, scatter, distribution, and diagnostic plots may use restricted ranges when this does not misrepresent magnitude;
- any break, normalization, log scale, or truncated range must be visible and explained.
6. Verify visually and programmatically
While the plotting code still holds the Matplotlib Figure, call visual_qa.audit_layout(fig) for glyph, clipping, and tick-overlap checks. Render a preview and inspect it against references/visual_review.md. The scripts/visual_qa.py <file> --preview ... CLI only renders an existing file; it does not run the Figure layout audit. Check clipping, overlap, missing glyphs, tick/label meaning, panel alignment, legend obstruction, grayscale interpretation, and whether the visual takeaway matches the source data.
Apply up to two bounded automatic repair passes for layout or rendering defects. If the scientific encoding or interpretation remains disputed, stop and report the decision rather than repeatedly restyling it.
7. Export and audit
Prefer vector output for line art and use the venue-required raster format and resolution for images. Run scripts/check_figure.py --strict for supported file checks. Strict exits 2 for FAIL and 3 for UNKNOWN (required checks not completed); exit 0 may include WARN and does not certify visual or full venue compliance. See references/publication_checklist.md for format coverage and export versus display dimensions. Re-check the figure at its actual manuscript display size after inclusion; scaling a vector graphic is allowed, but unreadable text or altered visual hierarchy is not.
Completion contract
A figure is complete only when:
- its Claim/Chain ID, source data/code, and manuscript role are recorded;
- its encoding, scale, uncertainty, exclusions, and caption are defensible;
- vector/raster output follows current venue requirements;
- programmatic and visual QA pass at final display size;
- the corresponding entry in
docs/10_figure_report.mdis updated.
References
| File | Open when |
references/chart_selection.md | Selecting a chart from variables, sample structure, and claim |
references/viz_pitfalls.md | Checking misleading or fragile encodings |
references/plot_recipes.md | Implementing the selected matplotlib/seaborn chart |
references/journal_specs.md | Establishing dimensions, fonts, DPI, and formats; verify current venue guidance before relying on presets |
references/data_profiling.md | Interpreting the profiling report |
references/visual_review.md | Perceptual QA of the rendered preview |
references/publication_checklist.md | Final formal checks |
Scripts
setup_style.py: configurable style and CJK font handlingprofile_data.py: CSV/XLSX/TSV data profilingvisual_qa.py: file preview CLI;audit_layout(fig)for in-memory Figure layout auditexport_figure.py: multi-format export and grayscale previewcheck_figure.py: strict output auditlayout_tools.py: reusable layout helpers
Dependencies
Use the existing project environment. See requirements.txt for core packages and optional features. XLSX profiling needs openpyxl; legacy XLS needs xlrd. PDF file audits need pypdf (or compatible PyPDF2); saved-PDF preview rendering needs PyMuPDF. Missing a dependency blocks only that feature and must be reported as incomplete, not passed. Do not install packages without the relevant authorization.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

