Create Word reports, memos, letters, and proposals with headings, tables, lists, and a table of contents. Also supports targeted edits to existing .docx files and reading or summarizing their text and tables.
Use this skill whenever the user asks for a Word document, a .docx file, or a report/memo/letter/proposal delivered as one — creating a new document from scratch, editing an existing .docx, or reading/summarizing the contents of one. Covers page layout, headings, tables, numbered/bulleted lists, page breaks, and a table of contents. Does not cover converting a .docx to PDF or .doc to .docx (see file-conversion for what conversions ARE supported in this environment).
Review the instructions the agent will receive before you install.
| Task | Approach |
|---|---|
| Create a new .docx from scratch | docx (docx-js) — write a TypeScript program (packages: ["docx"]) and run it via your coding tool |
| Edit an existing .docx | office-utils: unpack → edit XML → pack (see that skill) |
| Read / extract text from a .docx | python-docx or docx2python (Python, already installed — no pandoc needed) |
Always CREATE documents with docx (docx-js) in TypeScript — never with python-docx. python-docx is in this environment for reading existing documents only. Run the TypeScript program via whatever coding surface you have: the run_code tool directly (language="typescript", packages=["docx"]), or the coding_agent delegate — in that case restate these library/language requirements in its goal. The library is pre-installed in the standard sandbox image, so the install step is an instant no-op there.
docx-js (npm package docx) builds a document as a tree of JS objects, then serializes it. The hard-won pitfalls below are the difference between a .docx that opens cleanly in Word/Google Docs and one that shows a repair prompt or renders with the wrong layout — verify your generated file against every one of these before considering the task done.
{ width: 12240, height: 15840 } DXA. Always set Document.sections[0].properties.page.size explicitly — don't rely on the library default, which may not match what the user expects.WidthType.DXA, not WidthType.PERCENTAGE, for predictable rendering. Percentage widths compute inconsistently across renderers (Word vs. Google Docs vs. LibreOffice) when the table has merged cells or nested tables. Compute explicit DXA widths that sum to the usable page width (page width minus left+right margins).LevelFormat numbering config registered on the document, not literal unicode bullet characters (•, -) typed into a text run. A literal bullet character is not a list to Word — it can't be renumbered, indented, or styled as one. Define a numbering.config with LevelFormat.BULLET (or LevelFormat.DECIMAL for numbered lists) and reference it from each paragraph's numbering property.Paragraph, not a standalone element. Use new Paragraph({ children: [new PageBreak()] }) — inserting a bare PageBreak() outside a paragraph will fail to serialize correctly in some docx-js versions.TableOfContents) only picks up paragraphs styled with a real Heading style (HeadingLevel.HEADING_1 etc.), never plain bold text. If the user wants a working, clickable/updatable ToC, every section title must use a heading style — and the ToC itself needs a "right-click → Update Field" note in your response, since docx-js writes the ToC's field code but cannot pre-compute the resulting page numbers without a real layout engine.size (half-points) on any TextRun where the default (usually 11pt/22 half-points) doesn't match the user's request — half-points, not points: 12pt text is size: 24.Use the office-utils skill's unpack → edit → pack cycle: the document body lives in word/document.xml inside the unpacked tree. Locate the <w:p> (paragraph) or <w:tbl> (table) you need to change, make the smallest edit that achieves the goal, and repack. Do not attempt to regenerate the whole document via docx-js just to change one paragraph — you will lose any formatting/styles the original author applied that aren't represented in your regenerated version.
python-docx: Document(path).paragraphs / .tables for structured access (paragraph text + style, table cell grids).docx2python: better for a fast, structure-preserving plain-text dump when you just need the content, not to manipulate specific elements.Both are already installed in the sandbox — no install_packages call needed.
.doc → .docx conversion, and .docx → PDF rendering — both require LibreOffice, which is not installed in the sandbox yet. Tell the user this rather than attempting a workaround that will silently produce a wrong result.No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Build or extend a PipesHub connector, an integration that syncs content from an external service or database. The skill guides API research, design decisions, permission mapping, implementation, testing, and documentation of what was actually verified.
Load, clean, summarize, and join tabular data from CSV, Excel, JSON, or database exports using pandas. Check missing values, data types, and row counts, then independently verify key figures before reporting numbers printed by the code.
Turn data into charts suited to comparisons, trends, distributions, relationships, or shares. Add clear titles, units, and accessible colors, then export a static PNG or interactive HTML with hover details and zoom controls.
Choose suitable libraries for converting spreadsheet data and image files, or extracting text and tables from documents. Covers supported format pairs and identifies conversions unavailable in the described environment before work begins.
Unpack Word and PowerPoint files to inspect and edit their internal XML, the markup that stores document content. Repack targeted changes with XML syntax checks while preserving the original archive entry order.
Create PDF reports, invoices, and letters, extract text or tables from existing PDFs, and fill interactive form fields. The skill also guides table layout, text wrapping, and page breaks so generated documents fit their pages.