| name | matplotlib |
| description | Create static, highly customized figures for CS papers, benchmark reports, experiment analysis, and technical documentation. Use when default plotting tools are too limiting, when multi-panel layouts need precise control, or when exporting clean PNG, PDF, or SVG figures for research outputs. |
Matplotlib
Use this skill when you need exact control over figure structure, axes, annotations, and export settings.
Good Fits
- training curves and ablation figures
- resource tradeoff plots: latency, memory, throughput
- confusion matrices and calibration-style figures
- multi-panel paper figures
- custom benchmark visualizations that higher-level libraries do not support well
Local Resources
references/plot_types.md
references/styling_guide.md
references/common_issues.md
references/api_reference.md
scripts/plot_template.py
scripts/style_configurator.py
Working Rules
- Prefer the object-oriented API for nontrivial figures.
- Design the figure around the comparison the reader must notice first.
- Keep labels, legends, and tick density readable at paper scale.
- Use seaborn or plotly only when their defaults genuinely save time.
- Use
scientific-visualization when the task is primarily figure orchestration or publication styling across multiple plots.
Default Output
When the user asks for plotting help without more detail, return:
- figure plan
- recommended plot type and layout
- styling or export notes
- code skeleton if implementation is requested