pptx → high-fidelity PDF (Windows PowerPoint COM)

When to use

A user on a WSL machine wants a .pptx turned into a PDF and cares that it matches what PowerPoint shows. This is the default PDF-conversion path for PowerPoint files here. If the user only needs rough text extraction or a searchable/editable PDF, this is the wrong tool (see "Trade-off" below).

First-run setup

On first use, verify the environment and install dependencies — ask the user if anything is unclear:

1. WSL + Windows PowerPoint: powershell.exe -NoProfile -Command "$env:TEMP" must work, and Microsoft PowerPoint must be installed on the Windows side. If not, tell the user this skill requires WSL with desktop PowerPoint and stop.
2. Python deps: pip install Pillow img2pdf into whatever Python the user wants to use (ask once, remember their choice).

Why this approach (read once, it explains every step)

Three ways exist to make a PDF from a .pptx, and only one matches the screen:

1. LibreOffice --convert-to pdf — different layout engine, so it measures glyph widths slightly differently and re-flows the text: lines that fit on one line in PowerPoint wrap, shoving everything below them. Rejected.
2. PowerPoint's own vector PDF export — layout is perfect, text is selectable, but a PDF viewer re-rasterizes the font outlines with grayscale anti-aliasing at screen DPI, so the text looks heavier / "fatter" than PowerPoint's ClearType screen rendering. Fine for print, but the user noticed and disliked the on-screen difference.
3. Slide images → PDF (this skill) — PowerPoint itself rasterizes each slide to PNG, and those exact pixels go into the PDF. The viewer just shows pixels (no font re-rasterization), so it looks identical to PowerPoint at any zoom. The only cost: text is not selectable (it's an image).

The fonts come from PowerPoint itself, so nothing can "break" — there is no font substitution or missing-glyph tofu, regardless of what's installed in WSL.

How to run

Everything is in one script. Requires Pillow and img2pdf (pip install Pillow img2pdf):

python ~/.claude/skills/pptx-to-pdf/scripts/pptx_to_pdf.py "<path/to/deck.pptx>"

That produces <path/to/deck>.pdf — plain basename, no quality suffix (do not append _7K, _q95, _고화질; the user asked for clean names).

Defaults (what you get with no flags):

• Export every slide as PNG at 6827×3840 (7K) via Slide.Export(path,"PNG",w,h). Explicit pixel dimensions mean no Windows registry tweak is needed.
• Re-encode as JPEG quality 95, 4:4:4 (subsampling=0) — visually identical to lossless (measured max per-pixel diff ≤5/255; 0% of pixels differ by >5 levels) at ~⅓ the file size (e.g. ~127 MB vs ~362 MB for a 39-slide deck).
• Page size follows the slide aspect (16:9 → 13.333in × 7.5in; 4:3 → ×10in).

Flags (only when the user asks)

• --lossless — embed exact PNG pixels (byte-perfect, ~3× larger). Use only if the user explicitly wants a true-lossless/archival master.
• --quality N — JPEG quality (default 95). Lower (e.g. 92) for a smaller file.
• --width W --height H — export resolution. Drop to --width 3840 --height 2160 (4K) for a much smaller file that still prints at ~288 DPI.
• -o OUT.pdf — explicit output path.

Don't ask the user which quality each time — just run the default (7K q95). Only reach for flags when they request lossless, smaller, or a specific size.

Requirements / gotchas

• Windows PowerPoint must be installed and reachable via powershell.exe (checked during first-run setup). The script verifies the COM export succeeded and aborts loudly otherwise.
• The script stages the file under the Windows %TEMP% dir as ASCII input.pptx, because Korean filenames get mangled crossing the WSL↔Windows boundary (the original name is preserved for the output, only the staged copy is renamed). It auto-detects %TEMP% and cleans up after itself.
• A 39-slide deck takes ~1–3 min (PowerPoint renders each slide at 7K).
• Large input decks (200 MB+) and ~360 MB lossless outputs are normal; check free disk if doing many.

Trade-off to surface to the user

The output text is not selectable/searchable (it's slide images). If they need search or copy-paste, mention the options: use PowerPoint's vector PDF export instead (accepting the slightly heavier on-screen font), or add an invisible OCR text layer afterward with ocrmypdf (Korean OCR is imperfect).

Verify

After conversion, sanity-check by rendering a text-heavy slide and looking at it:

pdftoppm -f 2 -l 2 -r 110 -png "<out>.pdf" /tmp/_chk && # then view /tmp/_chk-02.png
pdfinfo "<out>.pdf" | grep -iE 'pages|page size'

Confirm page count matches the deck and a long title sits on the same line as in PowerPoint (proof that nothing re-flowed).