Document Export

Turn content into a real office file: a PDF, an Excel workbook, or a Word document. The skill ships three helper scripts that do the file writing deterministically; your job is to assemble the content and call the right one.

These scripts use the Python standard library only, so there is nothing to install for Excel and Word. PDF needs a browser or a PDF tool on the machine (the script finds it, or tells the user what to install).

Save every output to reports/ as YYYY-MM-DD-<slug>.<ext>, then tell the user the path.

PDF

PDF is produced by rendering an HTML file. Two cases:

• Rich, visual page (architecture, dashboard, comparison): use the visual-explainer skill to build the HTML, then convert it.
• Plain document (a letter, a report, a write-up): wrap the content in this minimal template and save it as an .html file:

<!DOCTYPE html>
<html><head><meta charset="utf-8"><title>TITLE</title>
<style>
  @page { margin: 2cm; }
  body { font: 11pt/1.55 -apple-system, "Segoe UI", Roboto, sans-serif;
         color: #1a1a1a; max-width: 46em; margin: 0 auto; }
  h1, h2, h3 { line-height: 1.25; }
  h1 { font-size: 1.8em; } h2 { font-size: 1.35em; margin-top: 1.6em; }
  table { border-collapse: collapse; width: 100%; }
  th, td { border: 1px solid #ccc; padding: 6px 10px; text-align: left; }
  code { background: #f4f4f4; padding: 1px 4px; border-radius: 3px; }
</style></head><body>
  <!-- CONTENT -->
</body></html>

Then convert:

python3 .claude/skills/document-export/to_pdf.py <input.html> reports/<YYYY-MM-DD>-<slug>.pdf

It tries headless Chrome first, then weasyprint, then wkhtmltopdf. If it reports that no engine is found, relay its install hint to the user; do not try to work around it.

Excel (.xlsx)

Build a JSON spec, write it to .tmp/export-spec.json (inside the kit, so saving it needs no permission prompt), and run the writer.

Spec shape:

{
  "sheets": [
    {
      "name": "Results",
      "header": true,
      "rows": [["Name", "Score"], ["Alice", 90], ["Bob", 85]],
      "widths": [24, 10]
    }
  ]
}

• header: true bolds the first row. widths (column widths in characters) is optional. Multiple sheets are allowed.
• Cell values may be strings, numbers, booleans, or null (an empty cell). Keep numbers as numbers in the JSON so Excel treats them as numeric.

Run it:

python3 .claude/skills/document-export/to_xlsx.py .tmp/export-spec.json reports/<YYYY-MM-DD>-<slug>.xlsx

You can also pipe the spec on stdin by passing - instead of a file path.

Word (.docx)

Build a JSON spec of blocks, then run the writer.

Spec shape:

{
  "title": "Optional document title",
  "blocks": [
    {"type": "heading", "level": 2, "text": "A section"},
    {"type": "paragraph", "text": "A plain paragraph."},
    {"type": "paragraph", "runs": [
      {"text": "mixed ", "bold": true},
      {"text": "formatting", "italic": true}
    ]},
    {"type": "bullets",  "items": ["first point", "second point"]},
    {"type": "numbered", "items": ["step one", "step two"]},
    {"type": "table", "header": true, "rows": [["Col A", "Col B"], ["1", "2"]]}
  ]
}

• heading levels are 1 to 3. A paragraph carries plain text (with optional bold / italic for the whole paragraph) or a list of runs for mixed formatting. table rows are lists of cell strings; header: true bolds the first row.

Run it:

python3 .claude/skills/document-export/to_docx.py .tmp/export-spec.json reports/<YYYY-MM-DD>-<slug>.docx

After exporting

• Leave .tmp/export-spec.json in place: .tmp/ is git-ignored scratch space and the next export overwrites it, so deleting it would only cost the user another permission prompt.
• Give the user the exact path of the file.
• If the user wants to share the file, that is an outward action: hand it over, do not send it anywhere yourself.

Scope

These writers cover the structure that carries content: values, headings, formatting, lists, simple tables, column widths. They do not do Excel formulas, charts, or merged cells, or Word images, footnotes, or page headers. If a request needs those, say so and either extend the script or, if the user has the openpyxl or python-docx library available, write a one-off script with it instead.