From f2de050587efc8a182a7423282f6dae7b3bf3d19 Mon Sep 17 00:00:00 2001 From: hrbrmstr Date: Sun, 17 May 2026 08:18:36 -0400 Subject: add: gribouille --- .../gribouille-skill/SKILL.md | 442 +++++++++++++++++++++ 1 file changed, 442 insertions(+) create mode 100644 2026/2026-05-17-gribouille/gribouille-skill/SKILL.md (limited to '2026/2026-05-17-gribouille/gribouille-skill/SKILL.md') diff --git a/2026/2026-05-17-gribouille/gribouille-skill/SKILL.md b/2026/2026-05-17-gribouille/gribouille-skill/SKILL.md new file mode 100644 index 0000000..9da974d --- /dev/null +++ b/2026/2026-05-17-gribouille/gribouille-skill/SKILL.md @@ -0,0 +1,442 @@ +--- +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: , + mapping: aes( + x: "", + y: "", + // add aesthetics as needed: + colour: "", + fill: "", + shape: "", + size: "", + alpha: "", + ), + 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 | -- cgit v1.2.3