@usejunior/docx-core

OOXML primitives and DOCX generation in TypeScript.

Install

npm install @usejunior/docx-core

Example

import { writeFile } from 'node:fs/promises';
import { generateDocx } from '@usejunior/docx-core';

const document = await generateDocx({
  sections: [{
    blocks: [{
      kind: 'paragraph',
      runs: [{ kind: 'text', text: 'Hello' }],
    }],
  }],
});

await writeFile('hello.docx', document);

DocumentSpec supports sections, headers and footers, fields, styles, tables, numbering, and comments. The supported runtime uses jszip and @xmldom/xmldom; it does not require Word, LibreOffice, Python, or .NET.

For two-document comparison, use @usejunior/docx-compare. For agent-driven editing, use @usejunior/safe-docx.

See the repository architecture, support contract, and package source and examples.

Bounded table-row editing

DocxDocument.insertTableRow(...) and deleteTableRow(...) edit existing body-level tables in clean or tracked mode. By default they accept only rectangular, unmerged tables with direct rows and cells; the experimental mergeAware: true option also admits validated horizontal gridSpan rows. Vertical merges, row offsets, nested tables, wrapped topology, table-property exceptions, and topology revisions still fail with a typed SafeDocxError.detail. Insertions return fresh paragraph anchors for chaining, and accept/reject resolve the emitted row revisions in body and supported side-story parts. For tracked edits, callers must seed RevisionContext.idState above existing package revision IDs; the MCP session layer performs that scan automatically.

Bounded table-column editing

DocxDocument.insertTableColumn(...) and deleteTableColumn(...) address zero-based logical tblGrid columns. Insertion requires a positive widthTwips and one explicit rowActions entry per row: { kind: 'cell', text } creates a physical cell, while { kind: 'growCell', side: 'left' | 'right' } widens the cell on that side of the boundary. These operations are clean-only and admit simple and horizontal-gridSpan rows. They reject tracked requests, vertical merges, offsets, nested tables, pending topology revisions, and deletion of cells with unresolved revisions. Invalid requests leave the document unchanged and report row/column/cell coordinates in SafeDocxError.detail; successful insertions return paragraph IDs for newly created cells. Column edits preserve live references to unaffected DOM elements.

Bounded table-cell editing

DocxDocument.splitTableCell({ anchorParagraphId, splitColumnIndex, existingSide, newCellText }) splits a horizontally spanning physical cell at the boundary before zero-based logical column splitColumnIndex, retaining its authored content on existingSide and returning the new cell's paragraph anchor. DocxDocument.absorbTableCell({ anchorParagraphId, absorbSide }) removes the anchored physical cell, discards its content, and expands the adjacent cell on absorbSide to cover the vacated grid slots. Neither changes tblGrid or other rows. A bookmark or comment range cut by removal loses its surviving partner endpoint even if that endpoint is in another cell or row. Both operations are clean-only and require direct body-level, non-vertically-merged tables with resolvable affected widths; percentage-width splits and percentage-width absorbing cells fail closed. Pending row/cell topology, revisions in a removed cell, nested tables, offsets, and tracked requests also fail before mutation. These APIs do not perform Word's shift-right/down insertion or shift-left/up deletion; those remain separate work under #1042.