@ai-gui/plugin-solid

Solid-geometry teaching figures for AIGUI. The model names a solid and the conditions on it — 正方体 ABCD-A1B1C1D1, 过 A、B1、D1 的截面 — and the figure is computed from those.

Install

pnpm add @ai-gui/plugin-solid

three is a dependency and is imported only when a figure is actually drawn, so a page whose answers contain no geometry never loads it.

Usage

import { solid } from "@ai-gui/plugin-solid"
import { buildSystemPrompt } from "@ai-gui/core"

<AIRenderer plugins={[solid()]} />

// The model has no prior for this block, so the spec is what makes it usable at all.
const system = buildSystemPrompt({ registry, plugins: [solid()], locale: "zh-CN" })

The model then emits:

平面 AB₁D₁ 截正方体，截面是正三角形。

```solid
{
  "solid": "cube",
  "label": "ABCD-A1B1C1D1",
  "edge": 2,
  "section": { "through": ["A", "B1", "D1"] },
  "highlight": [{ "plane": ["A", "B1", "D1"] }],
  "caption": "平面 AB1D1 截正方体所得的截面"
}
```

Why the protocol looks like this

Conditions, never results. The model says which three points the plane passes through; the section polygon is computed here. A model asked instead for the shape sometimes answers "五边形" when it is a hexagon, and a figure drawn from that answer would be wrong in a way a student cannot see. Vertex coordinates are refused for the same reason: they put the model's arithmetic into the picture.

The textbook's own vocabulary. A model is fluent in 正方体 ABCD-A1B1C1D1 — it has seen the notation hundreds of thousands of times — and is not fluent in {"vertices": [[0,0,0], …]}.

Every reference must resolve. A figure naming a point that was never defined is refused rather than drawn without it. {"plane": ["P", "A", "B"]} on a cone is the right shape in every field and still meaningless, because a cone has no A or B until onCircle puts them there.

Supported

| | | | --- | --- | | Solids | cube, cuboid, prism (3–6 sides), pyramid (3–6 sides), cylinder, cone, sphere | | Points | along a segment (on + at), a face centre (center), the foot of a perpendicular (foot), a point on a circle (onCircle + angle) | | Marks | segments (solid or dashed), section through three points, highlight of a line, plane, or angle | | Apex | apexOver: "A" puts the apex above one vertex, which is what makes PA ⊥ 底面 drawable — the default regular pyramid would contradict the explanation |

Not in this version, and the prompt spec tells the model to explain them in prose instead: net diagrams and shortest-path unfoldings, non-axial sections of cones and cylinders (they are conics, not polygons), animation of a moving point, compound or hollowed solids, coordinate annotation.

Options

• height?: number — figure height in CSS pixels, default 320.
• maxPoints?: number — refuse a figure introducing more than this many points, default 24.
• maxSourceBytes?: number — refuse a fence larger than this before parsing it, default 16 KiB.

Dragging turns the figure; there is no authoring, panning or unbounded zoom. theme is honoured, so a figure on a dark page is drawn for one.

Testing note

src/fixtures/ holds the 19 figures a model actually produced when given this plugin's prompt spec and twenty textbook questions. They are part of the test suite: a protocol change they stop parsing is a change that breaks answers already being written.

See the root README for the full plugin list.