path: root/2026/2026-05-17-gribouille/gribouille-skill/SKILL.md

---
name: gribouille
description: >
  Generate correct, idiomatic gribouille charts for Typst. Trigger on: "chart",
  "plot", "scatter", "histogram", "bar chart", "boxplot", "heatmap", "time series",
  or any request to visualise data in Typst using a Grammar of Graphics approach.
version: 0.2.0
min-binary-version: "4.0.0"
allowed-tools:
  - Read
  - Write
  - Edit
  - Bash
  - AskUserQuestion
---

# /gribouille

Generate correct, idiomatic [gribouille](https://m.canouil.dev/gribouille) charts for Typst.
Gribouille implements Wilkinson's Grammar of Graphics — the same mental model as ggplot2 and plotnine — compiled natively in Typst.

```
/gribouille scatter penguins flipper-len vs body-mass by species
/gribouille bar chart of mpg highway economy by class, faceted by cyl
/gribouille histogram of my-data.csv column "score"
/gribouille compose two panels: weight vs mpg and horsepower vs mpg
```

---

## Step 0 · Context detection

Determine two things silently before asking anything.

**A. Import mode:**

| Signal | Import line |
|---|---|
| User mentions Typst Universe / `@preview` | `#import "@preview/gribouille:0.0.1": *` |
| User has a local gribouille clone / `lib.typ` path | `#import "./path/to/lib.typ": *` |
| Ambiguous | Ask: "Are you using gribouille from Typst Universe (`@preview`) or a local clone?" |

**B. Embedding context:**

| Signal | Action |
|---|---|
| User is inside a typst-author document | Generate only the `#plot(...)` block; add Step 9 handoff note |
| User wants a standalone `.typ` file | Prepend `#set page(width: auto, height: auto, margin: 0.5cm)` |
| Unclear | Default to standalone |

---

## Step 1 · Intent extraction (silent checklist)

Before writing code, confirm these four dimensions. Ask only if 2+ are missing — one compact question at most, never a four-item form.

| Dimension | What to extract | Example |
|---|---|---|
| **Chart type** | What visual form? | scatter, bar, histogram, boxplot, line, heatmap |
| **Data** | Inline / built-in / external file? | `penguins`, a CSV path, or an array literal |
| **Variables** | Which columns → which aesthetics? | x=flipper-len, y=body-mass, colour=species |
| **Facets** | Split into panels? By what variable? | one panel per island |

If chart type and data are both clear from context, skip the question entirely.

---

## Step 2 · Select chart type

First-match table. Apply the first row that fits the user's description.

| User says / data shape | Geom combination | Notes |
|---|---|---|
| "scatter", "relationship", two continuous vars | `geom-point(size: 2pt)` | Add `geom-smooth(method: "lm")` if trend desired |
| "scatter + fit / regression / smoother" | `geom-point(size: 2pt, alpha: 0.65)` + `geom-smooth(method: "lm", se: true, alpha: 0.2)` | v1 supports `method: "lm"` only — not loess or glm |
| "scatter + convex hull / group outlines" | above + `geom-mark(method: "hull", expand: 5pt, alpha: 0.25)` | |
| "scatter + error bars" | `geom-point()` + `geom-errorbar(width: 0.3)` with `ymin`/`ymax` in mapping | |
| "line chart / time series / trend over x" | `geom-line()` | Sort data by x first; use `geom-area()` if area-under-line matters |
| "area chart" | `geom-area(alpha: 0.7)` | Requires ordered x |
| "bar chart (counts)" | `geom-bar()` | Uses stat-count by default; map `x` only |
| "bar chart (values / col chart)" | `geom-col()` | Data already has y values; map both `x` and `y` |
| "stacked bar" | `geom-bar(position: "stack")` | Add `fill` aesthetic |
| "dodged / grouped bar" | `geom-bar(position: "dodge")` | Add `fill` aesthetic |
| "filled / proportional bar (100% stacked)" | `geom-bar(position: "fill")` | y becomes proportion 0–1 |
| "histogram / distribution of continuous var" | `geom-histogram(bins: 12)` | Tune `bins` or use `binwidth` instead |
| "overlapping distributions / frequency polygon" | `geom-freqpoly(bins: 12)` | Line version of histogram; good for comparing groups |
| "boxplot / box-and-whisker" | `geom-boxplot()` | Add `geom-jitter(width: 0.2, alpha: 0.3)` for raw data overlay |
| "heatmap / tile map" | `geom-tile()` | Requires x, y, fill aesthetics |
| "2D histogram / bin2d" | `geom-bin-2d()` | Bins both x and y |
| "hex bin (dense scatter)" | `geom-hex()` | Alternative to scatter for large datasets |
| "contour lines" | `geom-contour()` | Requires x, y, z |
| "filled contours" | `geom-contour-filled()` | Requires x, y, z, fill |
| "error bars" | `geom-errorbar(width: 0.3)` with `ymin`/`ymax` in mapping | |
| "horizontal error bars" | `geom-errorbarh(height: 0.3)` with `xmin`/`xmax` in mapping | |
| "ribbon / confidence band" | `geom-ribbon(alpha: 0.2)` with `ymin`/`ymax` in mapping | Pair with `geom-line()` |
| "rug plot (marginal ticks)" | `geom-rug(sides: "bl")` | Stack with another geom |
| "path (connected in data order)" | `geom-path()` | Unlike `geom-line()`, does not sort by x |
| "step function" | `geom-step(direction: "hv")` | Options: "hv", "vh", "mid" |
| "segments / connectors" | `geom-segment(stroke: 1pt)` | `geom-curve()` for curved connectors |
| "text labels on data points" | `geom-text(mapping: aes(label: "col"))` | Combine with `geom-point()` |
| "labelled boxes (callouts)" | `geom-label(mapping: aes(label: "col"))` | Adds background box around text |
| "dot plot" | `geom-dotplot()` | |
| "Q-Q plot" | `geom-qq()` + `geom-qq-line()` | |
| "quantile regression" | `geom-quantile()` | |
| "spoke / wind vectors" | `geom-spoke()` | Requires x, y, angle, radius in mapping |
| "count overplotting" | `geom-count()` | Size encodes count |
| "function curve y=f(x)" | `geom-function(fun: x => calc.sin(x), n: 101)` | |
| "horizontal reference line" | `geom-hline(yintercept: 0)` | |
| "vertical reference line" | `geom-vline(xintercept: 0)` | |
| "diagonal / slope reference" | `geom-abline(slope: 1, intercept: 0)` | |
| "shaded rectangle region" | `geom-rect()` with `xmin`/`xmax`/`ymin`/`ymax` in mapping | |
| "polygon" | `geom-polygon()` | |
| "confidence ellipse" | `geom-ellipse()` | Or use `geom-mark(method: "ellipse")` |
| "faceted" | Any geom above + `facet: facet-wrap("col")` or `facet: facet-grid(rows: "r", columns: "c")` | |

Read `references/geom-table.md` for full parameter reference on any geom.

---

## Step 3 · Wire the data

Choose the pattern that matches the data source.

**Pattern A — Built-in dataset (no extra code needed):**
```typst
// penguins, mpg, and economics are exported by gribouille
#plot(
  data: penguins,   // or: mpg, economics
  ...
)
```

**Pattern B — Inline array literal:**
```typst
#let my-data = (
  (x: 1.0, y: 2.1, group: "a"),
  (x: 2.0, y: 3.0, group: "a"),
  (x: 3.5, y: 1.8, group: "b"),
)
#plot(
  data: my-data,
  ...
)
```

**Pattern C — External CSV (column headers become dictionary keys):**
```typst
#let raw = csv("data.csv", row-type: dictionary)
// Numeric values arrive as strings from csv(); cast as needed:
#let my-data = raw.map(r => (
  x: float(r.x),
  y: float(r.y),
  group: r.group,
))
#plot(
  data: my-data,
  ...
)
```

**Key rule — always quote column names in `aes()`:**
```typst
// CORRECT
mapping: aes(x: "flipper-len", y: "body-mass")

// WRONG — causes a compile error
mapping: aes(x: flipper-len, y: body-mass)
```

---

## Step 4 · Build the plot block

Canonical skeleton. Include only the arguments that are actually needed — omit optional sections when defaults suffice.

```typst
#import "@preview/gribouille:0.0.1": *

#set page(width: auto, height: auto, margin: 0.5cm)  // standalone only; remove in typst-author docs

#plot(
  data: <data>,
  mapping: aes(
    x: "<x-col>",
    y: "<y-col>",
    // add aesthetics as needed:
    colour: "<group-col>",
    fill: "<group-col>",
    shape: "<group-col>",
    size: "<numeric-col>",
    alpha: "<numeric-col>",
  ),
  layers: (
    geom-point(size: 2pt, alpha: 0.85),
    // stack additional geoms here in draw order (bottom to top)
  ),
  // scales: only include when overriding defaults
  scales: (
    scale-y-continuous(labels: format-comma()),  // for large numbers (>9999)
    // scale-x-log10(),
    // scale-colour-viridis-d(),
  ),
  labs: labs(
    title: "Chart Title",
    subtitle: "Optional subtitle",
    x: "X-Axis Label",
    y: "Y-Axis Label",
    colour: "Legend Title",
    fill: "Legend Title",
    caption: "Data source.",
  ),
  theme: theme-minimal(),  // alternatives: theme-classic(), theme-void()
  width: 12cm,
  height: 9cm,
)
```

**Scale auto-selection — apply these rules silently before emitting code:**

| Aesthetic + data type | Scale to emit |
|---|---|
| x/y continuous, values ≤9999 | Omit — default is fine |
| x/y continuous, values >9999 | `scale-x/y-continuous(labels: format-comma())` |
| x/y log scale requested | `scale-x/y-log10()` |
| x/y date values | `scale-x/y-date(date-format: "[year]-[month repr:numerical]")` |
| colour/fill discrete (categorical) | Omit — gribouille auto-trains; add `scale-colour-discrete(palette: ...)` only for custom colours |
| colour/fill continuous | `scale-colour-continuous()` or `scale-colour-gradient(low: rgb(...), high: rgb(...))` |
| colour viridis | `scale-colour-viridis-c()` (continuous) or `scale-colour-viridis-d()` (discrete) |
| colour ColorBrewer | `scale-colour-brewer(palette: "Blues")` |
| colour colourblind-safe | `scale-colour-okabe-ito()` |
| alpha mapped to data | `scale-alpha-continuous(range: (0.1, 1))` |

Read `references/scale-table.md` for full parameter reference.

**Categorical column rule — use `as-factor()` when a column holds numeric-looking strings used for grouping:**
```typst
// cyl column contains "4", "6", "8" as strings
mapping: aes(x: "displ", y: "hwy", colour: as-factor("cyl"))
//                                               ↑ forces discrete treatment
```

**Typst markup in labs — use the `typst()` helper, not raw strings:**
```typst
labs(
  title: typst("*Bold title* and #text(fill: blue)[blue text]"),
)
```

**Theme customisation — verified parameter names from source:**
```typst
theme: theme-minimal(
  text: element-text(family: "Source Sans Pro", size: 10pt),  // family:, NOT font:
  plot-title: element-text(size: 14pt, weight: "bold"),
  axis-text: element-text(size: 8pt),
  axis-line: element-line(thickness: 0.5pt),  // thickness:, NOT linewidth:
  tick-length: 0.08cm,
  tick-labels: true,
)
```

`element-text()` accepts: `size`, `weight`, `colour`, `angle`, `family`, `margin`
`element-line()` accepts: `colour`, `thickness`
`element-rect()` accepts: `fill`, `stroke`

---

## Step 5 · Faceting

Add a `facet:` argument to split the chart into small multiples.

**One variable → `facet-wrap`:**
```typst
facet: facet-wrap("island", ncolumn: 3),
// Options: nrow:, ncolumn:, scales: "fixed"|"free"|"free_x"|"free_y"
// labeller: label-value() (default) | label-both()
// axes: "margins" (default) | "all_x" | "all_y" | "all"
```

**Two variables → `facet-grid`:**
```typst
facet: facet-grid(rows: "sex", columns: "island"),
// v1 supports scales: "fixed" only
// Either rows: or columns: may be omitted (but not both)
```

---

## Step 6 · Multi-panel composition

Use `compose()` when the user wants multiple **independent** plots arranged together. Use `facet-wrap/grid` when it's the same chart split by a grouping variable.

```typst
#let p1 = plot(
  data: d,
  mapping: aes(x: "wt", y: "mpg", colour: as-factor("cyl")),
  layers: (geom-point(size: 2pt),),
  width: 6cm, height: 4cm,
  defer: true,  // REQUIRED for compose()
)
#let p2 = plot(
  data: d,
  mapping: aes(x: "hp", y: "mpg", colour: as-factor("cyl")),
  layers: (geom-point(size: 2pt),),
  width: 6cm, height: 4cm,
  defer: true,
)

#compose(
  p1, p2,
  layout: "grid",            // "grid" or "stack"
  columns: 2,
  collect: auto,             // auto hoists shared legends; none = keep per-panel
  guides-placement: "right", // "right" | "left" | "top" | "bottom"
)
```

---

## Step 7 · Guides / legend control

```typst
guides: guides(
  colour: guide-legend(position: "bottom", direction: "horizontal"),
  fill: guide-none(),
  shape: guide-legend(nrow: 1),
),
```

`guide-legend()` params: `title, nrow, ncolumn, reverse, position, direction, order, byrow`.
`position` accepts: `"top"`, `"right"`, `"bottom"`, `"left"`, `"none"`, a Typst alignment like `top + right`, or a dict `(dx:, dy:)`.

---

## Step 8 · Accessibility

Always emit an alt text string. Cache the plot in a `let` binding and call `get-alt-text()`:

```typst
#let p = plot(
  data: penguins,
  mapping: aes(x: "flipper-len", y: "body-mass", colour: "species"),
  layers: (geom-point(size: 2pt),),
  labs: labs(title: "Penguin Scatter", x: "Flipper Length (mm)", y: "Body Mass (g)"),
  theme: theme-minimal(),
  width: 12cm, height: 9cm,
)

#figure(
  p,
  caption: [Flipper length versus body mass for three penguin species.],
  alt: get-alt-text(p),
)
```

Tell the user to review and refine the generated alt text for their specific audience.

---

## Step 9 · Typst-author handoff

When the chart is destined for a full typst-author document (not a standalone file):

1. Generate the `#plot(...)` block without `#set page(...)`.
2. Tell the user:

> Paste the `#plot(...)` block inside a `#figure()` in your document:
>
> ```typst
> #let p = plot(
>   // ... your plot arguments ...
>   width: 12cm, height: 9cm,
> )
>
> #figure(
>   p,
>   caption: [Your caption here.],
>   alt: get-alt-text(p),
> )
> ```
>
> Remove `#set page(...)` if you added it during standalone testing — the enclosing typst-author document controls page geometry.

---

## Step 10 · Anti-patterns

Never emit these. Each has a correct alternative.

| Wrong | Correct | Why |
|---|---|---|
| `aes(x: col-name)` | `aes(x: "col-name")` | Column names must be quoted strings; unquoted is a Typst identifier |
| `aes(colour: "cyl")` when cyl holds `"4"/"6"/"8"` | `aes(colour: as-factor("cyl"))` | Numeric-looking strings need `as-factor()` for correct discrete training |
| `element-text(font: "MyFont")` | `element-text(family: "MyFont")` | The param is `family:`, not `font:` |
| `element-line(linewidth: 0.5pt)` | `element-line(thickness: 0.5pt)` | The param is `thickness:`, not `linewidth:` |
| `scale-y-continuous()` when y values are small | Omit the scale | Unnecessary boilerplate; defaults are fine |
| `#grid(columns: 2)[#p1][#p2]` for multi-panel | `#compose(p1, p2, layout: "grid", columns: 2)` | Typst `grid()` bypasses gribouille's legend hoisting |
| `labs(title: "*Bold*")` | `labs(title: typst("*Bold*"))` | Raw strings in labs are not parsed as Typst markup |
| `plot(..., data: csv("f.csv"))` | `#let d = csv("f.csv", row-type: dictionary)` then `plot(data: d, ...)` | `csv()` must use `row-type: dictionary` and be called outside `plot()` |
| `geom-histogram()` without an `x` mapping | `mapping: aes(x: "col")` always | Histogram requires a single `x` aesthetic |
| Omitting `width:` and `height:` | Always include `width: 12cm, height: 9cm` | Gribouille requires explicit dimensions; there is no default |
| `geom-smooth(method: "loess")` | `geom-smooth(method: "lm")` | v1 only supports `method: "lm"` — loess and glm are not available |
| `plot(..., defer: false)` inside `compose()` | `plot(..., defer: true)` | `compose()` only works with deferred specs |
| `scale-size-radius()` | `scale-radius()` | The export is `scale-radius`, not `scale-size-radius` |
| `scale-size-area(max.size: 6pt)` | `scale-size-area(range: (1pt, 12pt))` | The param is `range:`, not `max.size:` |
| `n.breaks:` in any scale | `n-breaks:` | Gribouille uses hyphens (Typst convention), not dots |

---

## Quick reference

**Three built-in datasets:**

| Symbol | Description | Key columns |
|---|---|---|
| `penguins` | Palmer Archipelago penguins | `flipper-len`, `body-mass`, `species`, `island`, `sex` |
| `mpg` | Fuel economy (ggplot2 mpg) | `displ`, `hwy`, `cty`, `class`, `cyl`, `drv` |
| `economics` | US economic time series | `date`, `unemploy`, `pop`, `psavert`, `uempmed` |

**Utility functions:**

| Function | Purpose |
|---|---|
| `as-factor("col")` | Force column to discrete treatment |
| `as-numeric("col")` | Force column to continuous |
| `typst("markup")` | Typst markup in labs fields |
| `after-scale(fn)` | Late-binding: evaluated after scale training |
| `after-stat(fn)` | Late-binding: evaluated after stat transform |
| `get-alt-text(p)` | Generate alt text from a plot result |
| `format-comma()` | Number formatter: 1,234,567 |
| `format-percent()` | Number formatter: 12.3% |
| `format-scientific()` | Number formatter: 1.23×10⁴ |
| `format-currency()` | Number formatter: $1,234 |
| `format-number(digits: N)` | Fixed decimal places |
| `format-wrap(width: N)` | Word-wrap labels at N characters |

**Three theme presets:**

| Theme | Character |
|---|---|
| `theme-minimal()` | No background panel, minimal chrome — default |
| `theme-classic()` | Classic R style with axis lines |
| `theme-void()` | No axes, no grid, just data |