PDF To Deck

Use this skill when the user uploads a PDF and wants a deck artifact derived from it.

Read ../../docs/input-contracts.md, ../../docs/quality-checks.md, and ../../docs/morph-integration.md before executing.

Use when

• The input is a binary PDF document.
• The expected output is a deck either summarizing or visually reconstructing the source.

Do not use when

• The input is plain markdown or a structured report (use deck-from-report).
• The user only wants a textual extract without a deck output.

Workflow

1. Classify the PDF

Inspect the PDF before extracting content. Decide between:

• summary distillation for structured PDFs with machine-readable text where the user wants a condensed or differently-shaped deck
• visual reconstruction for visually dense PDFs that must keep their original layout fidelity

Record the classification and the reason in the build note.

2. Extract content

For summary distillation, extract section headings, body text, callouts, and key tables. Preserve page citations so the deck can reference its source.

For visual reconstruction, capture page snapshots and structural metadata sufficient to rebuild the layout in deck form.

3. Pick the recipe and theme

Default to deck.report-summary for distillation. For reconstruction, choose a recipe that respects the original page sectioning and apply a theme close to the source palette unless the user requests a re-skin.

4. Choose the export mode

Default to pptx-static. Use pptx-morph only when narrative motion adds explanatory value over the original PDF.

5. Produce the artifact

• run pre-export QC, paying special attention to font fallback and image resolution
• write the export file plus build-notes.md, assets.json, and failures.json
• include page citations in the build note

Output

• a deck artifact in the requested mode
• a build note that records the classification decision, recipe, theme, and page citations
• an asset inventory and a failure list