docx

Core Rules

1. DXA Dual-Width Invariant for Tables: When generating Word documents using docx-js or OOXML, every table must declare columnWidths on the table element AND matching width on every individual TableCell using WidthType.DXA ($1\text{ in} = 1440\text{ DXA}$, $1\text{ pt} = 20\text{ DXA}$). Never use percentage widths, as they collapse in Google Docs and web office viewers.
2. Template Preservation via XML Unzipping: To modify existing styled documents or corporate templates, unzip the .docx archive, perform surgical edits in word/document.xml or word/styles.xml, and re-zip. Never attempt full-document regeneration with docx-js if preserving corporate typography and headers.
3. Dual-Pass Reading: When extracting or reviewing document content, convert .docx to Markdown via pandoc -f docx -t markdown for structural reading. For comment threads and tracked revisions, inspect word/comments.xml and word/document.xml (<w:ins>, <w:del>) directly.
4. Standard Page Geometry: Enforce A4 ($11906 \times 16838\text{ DXA}$) or US Letter ($12240 \times 15840\text{ DXA}$) with explicit $1\text{ in}$ ($1440\text{ DXA}$) margins on all section properties (<w:sectPr>).
5. Contract Manifest Verification: After creating or modifying a document, emit a valid document-manifest.json capturing file checksums, page estimates, and font configurations.

Suggested Process

1. Intake & Discovery: Determine whether the task requires creating a new document from scratch or modifying an existing .docx template.
2. Reading & Inspection: Convert existing files using pandoc -f docx -t markdown input.docx -o content.md to review prose structure.
3. Generation / Surgical Modification:
   • For scratch generation, author a Node.js script using docx (Packer.toBuffer).
   • For template editing, unpack via unzip input.docx -d unpacked/, modify unpacked/word/document.xml, verify XML well-formedness, and repack via zip -r output.docx *.
4. Table Formatting: Calculate column widths in DXA (e.g. 3-column table on US Letter: $4680, 2340, 2340\text{ DXA}$ summing to $9360\text{ DXA}$ printable width). Apply both table-level and cell-level widths.
5. Validation & Delivery: Inspect rendering via LibreOffice conversion to PDF, verify layout fidelity, and emit document-manifest.json.