sandbox-docx

Word documents in the sandbox

Specifier: @nodetool-ai/sandbox-docx. Declare it in the node's packages property and import it at the top of the body.

docx builds a document out of class instances (Paragraph, TextRun, Table, …) that cannot cross the guest boundary as plain data. This pack is a host module: you describe the document as a JSON element list and the host builds the real .docx bytes.

build — element list to .docx bytes

import { build } from "@nodetool-ai/sandbox-docx";

const bytes = await build({
  properties: { title: "Q3 Report", author: "NodeTool" },
  elements: [
    { type: "heading", text: "Q3 Report", level: 1 },
    { type: "paragraph", text: "Revenue grew 12% quarter over quarter.", bold: true },
    { type: "table", rows: [["Month", "Revenue"], ["Jul", "120k"], ["Aug", "134k"]] },
    { type: "pageBreak" },
    { type: "paragraph", text: "See appendix for methodology.", alignment: "CENTER" }
  ]
});
await workspace.writeBytes("report.docx", bytes);

Element types:

• heading — text, level (1–6, default 1)
• paragraph — text, alignment (LEFT/CENTER/RIGHT/JUSTIFY, default LEFT), bold, italic, fontSize (points, default 12)
• table — rows, a 2D array; every cell is stringified
• image — data (a Uint8Array), width/height in inches
• pageBreak — no fields

properties.title/author/subject/keywords set the document's metadata.

Gotchas

• build is async and returns a Uint8Array.
• Writing only. To read an existing .docx, use @nodetool-ai/sandbox-mammoth instead — these are two different libraries on two different sides of the format.
• Images are inches, not pixels. width/height follow docx's own convention; omit them and the image renders at a fixed placeholder size.
• 10 MB per image. The shared host-module input cap applies to each data field.