| name | scientific-artist |
| description | Generate scientifically accurate visual output from descriptions of scientific concepts or models. Use for sketches, figures, schematics, diagrams, publication graphics, Excalidraw drafts, Matplotlib figures, or TikZ output. |
Scientific Artist
Generate scientifically accurate visual output from a description of a scientific concept or model.
Workflow
- Read the user's description. Identify: equations, variables, spatial relationships, physical processes.
- Ask which tool stage to use (or infer from context — see selection table below).
- At publishing stage, disambiguate Matplotlib Advanced vs. TikZ if unclear.
- Generate output following the accuracy and style rules below.
- Present output for user feedback; iterate.
Tool selection
| User signal | Stage | Tool |
|---|
| "sketch", "draft", "quick diagram", "for a talk" | Drafting | Excalidraw MCP |
| "figure", "prototype", "next to my code", "iterate" | Prototyping | Matplotlib |
| "paper", "journal", "publication", "submit" | Publishing | Matplotlib Advanced or TikZ — ask if unclear |
| Ambiguous | — | Ask the user |
Scientific accuracy rules
These are non-negotiable across all stages:
- All labels and annotations use LaTeX math symbols.
- Model parameters are named constants at the top of the script, never magic numbers.
- Equations must match the actual model: correct variable names, subscripts, operator precedence.
- Arrow directions and spatial layout reflect physical meaning (gradients toward sources, torque arcs show rotation, sensing radii to scale).
- At publication stage: parameters from simulation output, not hardcoded.
Stage-specific guidance
Excalidraw (drafting)
Use the Excalidraw MCP tools (create_view, save_checkpoint). Call read_me before first use.
- Correct topology: right components connected to right components.
- Proper variable names on all labels (e.g., "Cell i", not "Cell 1").
- Physically meaningful spatial arrangement from the start.
- No equations required, but variable names must be correct.
- See
references/excalidraw-patterns.md for MCP usage and layout conventions.
Matplotlib (prototyping)
- Enable full LaTeX:
matplotlib.rcParams['text.usetex'] = True and font.family = 'serif'.
- Parameters as named constants at script top.
- Plot functions accept
ax argument (project convention).
- Save as PDF/SVG to
figures/ or outputs/.
- See
references/matplotlib-patterns.md for style patterns and helpers.
Matplotlib Advanced (publishing — Python pipeline)
- GridSpec multi-panel layouts with labeled panels (A, B, C...).
- Simulation-coupled panels: run real physics to generate data (Bessel functions, Monte Carlo, phase diagrams).
- Comparison tables via FancyBboxPatch.
panel_bg helper for consistent panel styling.- See
references/matplotlib-advanced-patterns.md for multi-panel and simulation patterns.
TikZ (publishing — LaTeX)
- Native LaTeX math via amsmath/amssymb.
- Reusable
\tikzset styles for cells, boxes, arrows.
- Relative positioning via TikZ libraries.
- Compile with
pdflatex for vector PDF.
- See
references/tikz-patterns.md for style definitions and patterns.
Examples
Working examples in examples/:
| File | What it demonstrates |
|---|
schematic-matplotlib.py | Single-panel schematic with LaTeX math, named parameters, color palette |
schematic-advanced-matplotlib.py | 6-panel simulation-coupled figure: K0 field, 2D heatmap, sine torque, heading update, Monte Carlo, phase diagram, comparison table |
chemoattraction-model-plan.md | Example "plan" input — the physics document that drove the advanced figure |
schematic-tikz.tex | Publication-ready TikZ schematic with reusable styles |
