docs.write_docx

Schema

Input = {
  path: string,                // .docx appended automatically if missing
  title: string,
  markdown: string,             // headings, paragraphs, lists, bold/italic, code, tables
  subtitle?: string,
  author?: string,
}

Output = {
  path: string,
  size: number,
  mime: 'application/vnd.openxmlformats-officedocument.wordprocessingml.document',
  blocks?: number,
}

How the document is built

markdownToDocx() parses the markdown with a small hand-rolled block parser (packages/skills/src/docs/markdown.ts) into headings, paragraphs, quotes, rules, fenced code, ordered/unordered lists, and tables, then renders each block type with the docx library into real Word structures — not an HTML-to-Word conversion. Headings use a serif display font (Georgia), body text uses Calibri, code gets a monospace font with a light shaded background, and tables get a shaded header row. A title page block (title + optional subtitle) is always prepended. This is the engine behind the Business plan and Research report generators, both of which declare docx as an output kind.

Example

curl -s -X POST http://localhost:4000/api/studio/skills/invoke \
  -H 'content-type: application/json' -b cookies.txt \
  -d '{"name":"docs.write_docx","input":{"path":"plan","title":"Lumen Bakehouse","subtitle":"Business plan","markdown":"## Summary\n\nA neighbourhood sourdough bakery..."}}'

{ "ok": true, "output": { "path": "plan.docx", "size": 14832, "mime": "application/vnd.openxmlformats-officedocument.wordprocessingml.document", "blocks": 2 } }

A sibling not in the canonical list

docs.write_pdf (same input shape, minus author) exists alongside it, rendering the identical markdown parse tree onto real PDF pages with pdf-lib — page breaks, wrapped text, code blocks with a shaded box, and simple tables — but it is not one of the 25 names in BUILTIN_SKILLS, so no generator can declare it as an allowed skill yet even though it’s registered and invokable.