| name | scientific-visualization |
| description | Orchestrate publication-ready figures for CS papers, posters, slides, and benchmark reports. Use when assembling multi-panel result figures, harmonizing styles across plots, preparing export settings, or turning experiment outputs into clear visual evidence for submissions and technical communication. |
Scientific Visualization
Use this skill when figure quality, consistency, and evidence communication matter more than raw plotting mechanics.
Typical CS Outputs
- benchmark comparison figures
- ablation summaries
- latency / memory / accuracy tradeoff panels
- error analysis views
- embedding or representation comparison panels
- reproducibility or scaling plots
- poster or slide figure sets with consistent styling
Workflow
- Identify the claim each figure must support.
- Choose the smallest set of plots that makes the comparison interpretable.
- Standardize colors, labels, scales, and panel conventions.
- Export in formats appropriate to the destination: paper, slide, poster, or README.
Local Resources
Assets:
assets/publication.mplstyleassets/presentation.mplstyleassets/color_palettes.pyassets/nature.mplstyle as an inherited legacy style, not the default
References:
references/publication_guidelines.mdreferences/color_palettes.mdreferences/matplotlib_examples.mdreferences/journal_requirements.md only when a venue genuinely requires it
Scripts:
scripts/figure_export.pyscripts/style_presets.py
Working Rules
- Optimize first for truthful comparison, then for polish.
- Use consistent semantics across panels.
- Label uncertainty clearly when seeds, folds, or repeated trials matter.
- Do not overload a paper figure with dashboard-level detail.
- If the figure is exploratory rather than final, use seaborn or plotly directly instead.
Default Output
When the user asks for figure help without more detail, return:
- figure set recommendation
- panel structure and style notes
- export guidance for the target medium
- risks that could make the visual misleading
