figure

Files (44)

• LICENSE
• README.md
• SKILL.md
• openai.yaml · agents/
• design-decisions.md · docs/
• nature-science-style-guide.md · docs/
• README.md · examples/01-line-plot/
• README.md · examples/02-bar-ablation/
• README.md · examples/03-architecture-diagram/
• README.md · examples/04-venue-switch/
• README.md · examples/05-defect-fix/
• demo-venue-switch.py · examples/live-demos/
• pdf_to_png.py · examples/live-demos/
• verify_v2.py · examples/live-demos/
• SKILL.md · skills/diagram-drawio/
• drawio-template.xml · skills/diagram-drawio/references/
• SKILL.md · skills/diagram-tikz/
• templates.tex · skills/diagram-tikz/references/
• SKILL.md · skills/figure-anti-defect/
• fix_figure.py · skills/figure-anti-defect/
• lint_figure.py · skills/figure-anti-defect/
• defect-catalog.md · skills/figure-anti-defect/references/
• SKILL.md · skills/plot-mpl/
• figure_venue.py · skills/plot-mpl/
• plot_mpl.py · skills/plot-mpl/
• figure-contract.md · skills/plot-mpl/references/
• figure-taxonomy.md · skills/plot-mpl/references/
• nature-nmi-visual-preset.md · skills/plot-mpl/references/
• palette-okabe-ito.md · skills/plot-mpl/references/
• venue-widths.md · skills/plot-mpl/references/
• bar.py · skills/plot-mpl/templates/
• heatmap.py · skills/plot-mpl/templates/
• image_grid.py · skills/plot-mpl/templates/
• line.py · skills/plot-mpl/templates/
• multi_panel.py · skills/plot-mpl/templates/
• scatter.py · skills/plot-mpl/templates/
• violin.py · skills/plot-mpl/templates/
• SKILL.md · skills/venue-switch/
• 01-neurips-cvpr.md · skills/venue-switch/references/
• 02-nature.md · skills/venue-switch/references/
• 03-ieee-tpami.md · skills/venue-switch/references/
• 04-academic-classic.md · skills/venue-switch/references/
• venue_switch.py · skills/venue-switch/
• test_anti_defect.py · tests/

SKILL.md (first lines)

# Figure — Router (v0.2, Nature-compatible)

You are a figure skill router for academic papers. Pick the right sub-skill, read its `SKILL.md`, and execute that workflow inline.

Example prompts: "plot a line chart of X", "fit this figure to Nature", "Nature figure".

v0.2 changes:
- All 5 chart templates (line/bar/scatter/heatmap/violin) now produce Nature-style figures:
  editable SVG output, three-font Arial/DejaVu/Liberation stack, `svg.fonttype='none'`,
  right + top spines off, legend frame off, error bars, "ours" highlighted in red.
- Added 2 new templates: `multi_panel.py` (4-panel GridSpec for Figure-1 style figures)
  and `image_grid.py` (attention/segmentation samples).
- Added `figure_venue.py`: a single entry point `apply_venue(venue, backend='python')` that
  configures matplotlib rcParams for 6 supported venues including `nature_nmi` (Nature
  Machine Intelligence pastel multi-panel style).
- Added `plot_mpl.py` CLI: `python plot_mpl.py line --x "..." --y "..." --venue nature --out fig.svg`.
- Added `venue_switch.py` CLI: `python venue_switch.py neurips out.pdf`.
- Added `lint_figure.py` that actually implements the 15-item defect catalog
  (was previously only 5 file-level checks). Returns P0/P1/P2 findings.
- Added `fix_figure.py` that auto-patches 7 common P0 defects (font size, color names,
  rotation, units, missing labels, tight bbox, close fig).

These match the production standard of `Yuan1z0825/nature-skills/nature-figure` (the
figures4papers-based NMI-style figure skill): same hard rules (svg.fonttype='none',
Arial/DejaVu/Liberation, spines off, frameon=False), same NMI pastel palette.

## What you route to

Five sub-skills live in `./skills/`. Pick the closest match:

| Sub-skill | When | Read |
|---|---|---|
| `plot-mpl` | User has data (numbers, arrays, CSV) and wants a matplotlib data plot (line / bar / scatter / heatmap / violin / multi-panel / image-grid) | `./skills/plot-mpl/SKILL.md` |
| `diagram-tikz` | User wants a structural diagram (architecture, pipeline, geometry) embeddable in LaTeX as TikZ | `./skills/diagram-tikz/SKILL.md` |
| `diagram-drawio` | User wants a roadmap / graphical abstract / free-form layout best done in draw.io | `./skills/diagram-drawio/SKILL.md` |
| `venue-switch` | User has an existing figure and wants to resize / recolor / restyle for a specific venue (NeurIPS / Nature / IEEE / NMI / CVPR) | `./skills/venue-switch/SKILL.md` |
| `figure-anti-defect` | User has a figure (PNG/SVG/PDF) or .py script and wants a defect check (legend overlap, font size, tick collision, color-blindness, missing units, Y not 0, etc.) | `./skills/figure-anti-defect/SKILL.md` |

## Nature/NMI visual preset

When the user says "Nature figure", "NMI", "Nature Machine Intelligence", "Nature Methods", "high-impact journal figure", "publication-grade multi-panel", or "make it look like Nature", route to `plot-mpl` or `venue-switch` and require `plot-mpl/references/nature-nmi-visual-preset.md`.

## Figure Contract first

For any new data figure, especially Nature/NMI, IEEE, multi-panel, benchmark, ablation, or reviewer-facing figures, require `plot-mpl/references/figure-contract.md` before code generation. The agent must draft a short Figure Contract covering the core claim, data source, data units, transform/statistics, panel map, venue, visual priority, and reviewer risk.

If any data meaning, column mapping, grouping, metric, unit, baseline, seed/run aggregation, error-bar definition, transform, or panel layout is uncertain, ask the user to confirm the Figure Contract before routing into plotting. Do not treat this as a style-only approval; the confirmation is mainly to ensure the next step uses the correct data and scientific intention.

## How to route

1. Read the user's request. Most requests fit one sub-skill cleanly.
2. If ambiguous, ask 1-2 questions before routing. The five question patterns are at the bottom of this file.
3. For new data plots, read `plot-mpl/references/figure-contract.md` and draft/confirm the Figure Contract before plotting unless the user's command is already complete and unambiguous.
4. Read the sub-skill's SKILL.md before doing anything. The sub-skill defines what input to expect, what to ask the user, and what to output.
5. Continue with that sub-skill inline. Do not stop after announcing the route.
6. Never bypass a sub-skill and generate the figure yourself. The sub-skills contain the templates, scripts, and style rules.

## What you never do

- You don't generate matplotlib/TikZ/drawio code before reading the routed sub-skill.
- You don't decide venue style — `venue-switch` reads the 4 venue rule docs and `figure_venue.apply_venue` enforces them.
- You don't grade figures without `figure-anti-defect`'s defect catalog.
- You don't say "looks fine" without running the sub-skill.

## Output format (every sub-skill should follow)

```
### Input
<what user gave>

### Decision
<which sub-skill ran, and why>

### Output
<matplotlib script / TikZ code / drawio XML / recolored PNG / defect report>

### Why
- rule 1 (source: ...)
- rule 2 (source: ...)

### Trade-offs
- what I didn't do and why
- what the user might want to overrule
```

## 5 ambiguous-question patterns

Use these to clarify before routing:

1. "Do you have **data** (numbers), or are you describing a **diagram** in words?"
2. "Will the figure go in **LaTeX** (use TikZ) or **Word / web** (use drawio or PNG)?"
3. "What's the **target venue** (NeurIPS / Nature / IEEE / CVPR / no specific)?" — for "Nature" or "high-impact", use `venue='nature_nmi'` to get the multi-panel pastel style.
4. "Do you have an **existing figure** to modify, or are you creating from scratch?"
5. "Will you **compile the figure yourself** (TikZ needs LaTeX) or do you need a PNG/SVG output?"

## Reference data (shared by all sub-skills)

- `skills/plot-mpl/references/figure-taxonomy.md` — chart type decision tree
- `skills/plot-mpl/references/figure-contract.md` — contract-first data, panel, transform, venue, and reviewer-risk confirmation before plotting
- `skills/plot-mpl/references/palette-okabe-ito.md` — color-blind safe palettes
- `skills/plot-mpl/references/venue-widths.md` — venue column widths and font sizes
- `skills/plot-mpl/references/nature-nmi-visual-preset.md` — Nature/NMI visual preset for editable SVG, panel labels, and low-saturation styling
- `skills/figure-anti-defect/references/defect-catalog.md` — defect checks

Use packaged references first. If the user provides external exemplar figures, treat them as optional visual guidance rather than a required install dependency.

## Example invocations

- "plot a line chart of X" → `plot-mpl` (5 templates + multi_panel + image_grid)
- "画个架构图" → `diagram-tikz` if LaTeX, `diagram-drawio` if Word/web
- "fit this figure to Nature" → `venue-switch` (then ask: column width? or single column?)
- "make a Nature NMI multi-panel figure" → `plot-mpl` with `multi_panel.py` + `venue='nature_nmi'`
- "check this figure" → `figure-anti-defect` (15-item catalog now actually runs)
- "fix this figure's defects" → `figure-anti-defect/fix_figure.py`
- "改这个图成 NeurIPS 风格" → `venue-switch` to neurips-cvpr
- "make a graphical abstract" → `diagram-drawio` (free-form, not strict LaTeX)
- "bar chart of X vs Y" → `plot-mpl` with `bar.py` template or `plot_mpl.py bar --data-csv`

From RTCartist/paper-suite, MIT.

More from RTCartist/paper-suite