0.2.0

Excel workbooks convert in full.

Hidden sheets were being skipped on the assumption that hiding one says the
data doesn't matter. It doesn't — hiding is a presentation choice, and a
workbook's hidden sheets are as often the raw table a visible pivot summarises
as they are scratch space. Skipping them quietly dropped the data the note was
about.

• Every sheet is converted, hidden ones included. Convert hidden sheets
  in settings turns that off.
• Skipped sheets are named, not counted. Skipped (hidden): raw data (monthly), Region, Agebracket instead of 18 hidden sheets skipped — a
  count tells a reader nothing they can act on. Empty sheets, and sheets whose
  worksheet part is missing from the file, are named the same way; the latter
  used to vanish without a trace.
• Coverage is in the frontmatter. sheets_converted: 23/23 at the top of
  the note, so how much of the workbook made it across is the first thing you
  see rather than a footnote after two thousand lines.
• Load-bearing hidden sheets are kept regardless of the setting. If a
  visible sheet's formulas, a pivot cache, a chart series or a workbook-scoped
  named range reads from a hidden sheet, it is converted and reported. That
  dependency is a far stronger signal than the hidden flag.

Nothing changed in how cells are read: dates still come out as dates,
percentages as percentages, and 27.38 still isn't 27.383982300884924.

Installing

Copy main.js and manifest.json into
<vault>/.obsidian/plugins/convert-to-markdown/, then enable Convert to
Markdown under Community plugins. Existing installs update from within
Obsidian.

Desktop only.

0.1.0

First release.

Converts PDF, Word, PowerPoint, Excel and image files into Markdown notes,
with their images. No Python, no pip, no cloud service — nothing you convert
leaves your machine.

Each format is read through its own structure rather than one generic
abstraction, which is what keeps the output clean:

• Word — heading levels from Word's own styles, ordered vs. bullet lists
  resolved through numbering.xml, tables, hyperlinks, bold/italic
• PowerPoint — one section per slide in presentation order, title
  placeholders as headings, speaker notes, slide tables
• Excel — dates instead of serial numbers, 27.38 instead of
  27.383982300884924, hidden helper sheets skipped
• PDF — multi-column layouts read in order, headings from font size,
  paragraphs rejoined and de-hyphenated, running headers and footers dropped
• Images — local OCR via Tesseract

Images are pulled out of every format and embedded where they sat — a figure
inside a Word table cell lands in that table cell. Identical images are
written once.

Installing

Copy main.js and manifest.json into
<vault>/.obsidian/plugins/convert-to-markdown/, then enable Convert to
Markdown under Community plugins.

Desktop only.

Note on OCR

Converting an image file downloads the recognition engine and English model
(~9 MB) on first use, then caches it. The image itself is never uploaded. On
machines where that CDN is blocked, point the OCR engine folder setting at
a vault folder holding the two files and OCR makes no network request at all.

Known limits

• Scanned PDFs are reported as an error rather than OCR'd
• Column detection is geometric, so a layout with no gutter running through it
  can still come out misordered
• Excel formulas export as their last-calculated value