Tool used: Mammoth (mwilliamson/mammoth.js).

This skill is for an agent that receives Microsoft Word .docx files and needs to convert them into clean HTML for a publishing or content-ingestion pipeline. The agent uses Mammoth to extract headings, lists, links, images, and other semantic structure from the source document, then hands the resulting HTML to the next system in the workflow, such as a CMS importer, a markdown cleanup step, or a review queue. The job-to-be-done is not “use a document converter.” It is “take an authored Word document and turn it into web-ready markup with minimal presentation junk.”

Invoke this when the user already has content in .docx and wants an automation step that converts it into clean output for websites, knowledge bases, newsletters, or editorial review. Do not invoke it just because Mammoth exists as a package. Use it when the agent’s job is specifically to ingest .docx content, preserve semantic meaning where possible, and feed that output into a broader automation. If the user needs round-trip editing, pixel-perfect rendering, or broad format conversion beyond Word-to-HTML, this is out of scope.

The boundary matters. Mammoth is not being listed here as a generic library card. The skill is narrowly about document-ingestion behavior: accept a .docx, convert it to simple HTML, note that complex formatting may not survive perfectly, and pass the normalized result onward. Typical integration points include Node.js automation, CMS publishing flows, static site generators, editorial pipelines, and post-processing steps that sanitize or transform the produced HTML.