PowerPoint Construction

Create the Deck

This step applies only when talk.engine is pptx (or null with a pptx inference confirmed in Branch on the Engine).

Read the template path from speaker-profile.json → infrastructure.template_pptx_path. The deck is built by the real PowerPoint app from a flat op sequence: BuildDeck opens a uniquely-named COPY of the template (for its custom layouts + masters), deletes the template's demo slides, then creates every slide from the ops and saves the output. You emit the ops while walking the outline (Walk the Outline below), then validate and build:

"{python_path}" "{speaker_toolkit_root}/skills/presentation-creator/scripts/validate-deckops.py" ops.txt
bash "{speaker_toolkit_root}/skills/presentation-creator/scripts/build-deck.sh" '{template_copy_pptx_path}' '{output_path}' ops.txt

Op vocabulary, field layout, and state rules: deckops-spec.md. The template is read-only — pass a uniquely-named copy. macOS + Microsoft PowerPoint only. Before the first pass run "{python_path}" "{speaker_toolkit_root}/skills/presentation-creator/scripts/deckops-doctor.py" --vault-root "{vault_root}" and require status: ok; anything else routes into deck-editing-setup.md at the step its next_step names. Do not guess whether this is a first run — the doctor answers it, and it also catches a closed container or a stale macro import before a long build half-finishes.

Illustrations (when illustration strategy is defined)

If the outline includes an style_anchor block, illustration generation, build generation, and deck application are owned by the illustrations skill. Build the deck structure first (Walk the Outline through Present to Author below), then delegate:

Skill(skill: "illustrations")

The skill generates missing illustrations, generates build sequences for any slide with a builds[] block, and runs apply-illustrations-to-deck.py to swap images into the deck, reposition titles into Safe zones, position IMG+TXT image+text columns, and insert build sequences. Returns control here once images are approved and applied.

For non-illustrated slides and EXCEPTION-format slides, handle inline as normal — the [IMAGE NN] placeholder resolves to a real asset that presentation-builder inserts during the slide walk.

Which ops to omit on illustrated slides

For FULL and IMG+TXT slides, emit only the slide structure (layout, TITLE, FOOTER) and omit the IMAGE op — the slide is left without a picture shape. The illustrations skill fills it in the post-build apply pass: FULL slides get a slide BACKGROUND FILL (set by the PowerPoint apply-backgrounds.sh pass, so the layout's halftone-dot overlay covers them); IMG+TXT slides get a left-column picture shape via apply-illustrations-to-deck.py.

When outline.yaml's style_anchor.composition is poster-theatrical, also omit the TITLE and FOOTER ops for the illustrated FULL slides (those with an image_prompt) — the title and footer are rendered into the illustration itself, so the only post-build inserts on those slides are the background fill and the QR code. A FULL slide without an image_prompt gets no illustration, so it keeps its TITLE and FOOTER ops.

The illustrations skill reads outline.yaml directly (style_anchor plus per-slide image_prompt / builds) — no surfacing or format translation needed.

Post-build pass order

The passes below renumber or overwrite each other, so run them in exactly this order:

1. Expand builds — if any slide has progressive-reveal builds, expand them FIRST. Emit the manifest and the expanded notes map in one call, passing the pre-expansion notes JSON (see Inject Speaker Notes below for its format):

   "{python_path}" "{speaker_toolkit_root}/skills/illustrations/scripts/build-expansion-manifest.py" \
       outline.yaml illustrations/builds/ --out builds-manifest.json \
       --notes notes.json --notes-out notes-expanded.json
   bash "{speaker_toolkit_root}/skills/presentation-creator/scripts/expand-builds.sh" \
       '{deck_copy_pptx}' '{expanded_pptx}' builds-manifest.json

   Expansion replaces each parent slide with its build frames as full-bleed slides and renumbers every later slide, so notes, backgrounds, and QR key on the POST-expansion deck. notes-expanded.json is already keyed that way: each parent's note is split at its [BUILD NN] cue lines, one segment per frame (per skills/illustrations/references/builds.md), and every other note is moved past the inserted frames. A refusal names the slide and the cue to fix; a WARN on stderr means a build slide's whole note went on its first frame because it has no cue.

2. Inject speaker notes — Inject Speaker Notes below. When the deck was expanded, inject notes-expanded.json into the expanded deck exactly as written; never edit or re-key it by hand. With no builds, inject notes.json directly.

3. Apply backgrounds — set the FULL-slide backgrounds via "{speaker_toolkit_root}/skills/presentation-creator/scripts/apply-backgrounds.sh" using the manifest from the apply pass. It must run LAST; any later python-pptx save would re-drop the per-slide background fills.

See rules/deck-editing-rules.md.

Template Layout Map

Read the layout map from speaker-profile.json → infrastructure.template_layouts[].

Each layout entry has: index, name, placeholders[], and use_for.

Layout Selection Logic (generic mapping)

Illustrated slides (FULL / IMG+TXT): the illustrations skill owns layout choice and positioning — this skill leaves the slide structure with a title + empty body and the illustrations skill applies the image post-walk. See the illustrations skill's skills/illustrations/references/generation.md and skills/illustrations/references/builds.md for the format vocabulary, geometry, and build-insertion rules.

EXCEPTION format: real assets only — pick the layout per content type (bullet list, comparison, etc.) and resolve the image source from the [IMAGE NN] placeholder, not from illustrations/.

Also review the template catalog in slide-design-spec.md Section 7 for rich patterns (SWOT, timelines, funnels, competitor matrices, etc.) that may be available in the speaker's template.

Walk the Outline — Slide Generation Workflow

Walk the outline in order and append ops to the sequence — one slide is a SLIDE op followed by its content ops. Field layout and geometry (points): deckops-spec.md.

1. Start the slide with the right layout (0-based index from profile template_layouts): SLIDE␟<layout_index>

2. Set placeholders — title, subtitle, body: TITLE␟<text> · SUBTITLE␟<text> · BODY␟<text>

3. Bullet list content — one op per bullet, indent level 0-based: BULLET␟<level>␟<text>

4. Image/meme slides — title-only or blank layout, then an image: IMAGE␟<left>␟<top>␟<width>␟<height>␟<path>

4b. For illustrated slides (Format: FULL or IMG+TXT):

Emit the slide structure (layout, TITLE, FOOTER) and OMIT the IMAGE op. The illustrations skill applies the image after the build completes — see the Illustrations section above.

For EXCEPTION format:

• Use appropriate layout for the content type (bullet list, comparison, etc.)
• Image source comes from the [IMAGE NN] placeholder, not from illustrations/

5. Non-placeholder text (captions, annotations): TEXT␟<left>␟<top>␟<width>␟<height>␟<text>

6. Shapes (dividers, accent boxes): SHAPE␟<msoAutoShapeType>␟<left>␟<top>␟<width>␟<height>

7. Tables and charts: TABLE␟<rows>␟<cols>␟<l>␟<t>␟<w>␟<h> then CELL␟<row>␟<col>␟<text> per cell; CHART␟<xlChartType>␟<l>␟<t>␟<w>␟<h> then CAT␟<name> and SERIES␟<name>␟<v>␟<v>….

Background Colors

Read design_rules.background_color_strategy from the speaker profile, then emit a BG␟<r>␟<g>␟<b> op on the slide. Common strategies:

• random_non_repeating — pick a random saturated color, never repeat on adjacent slides
• theme_sequence — follow the template's built-in color rotation

Read design_rules.white_black_reserved_for to know when white/black backgrounds are appropriate (typically full-bleed image/meme slides only).

Footer

Read the footer text pattern from design_rules.footer.pattern, substitute conference-specific values, and add a footer to EVERY slide with a FOOTER␟<text> op. The op carries only the text — BuildDeck applies fixed footer geometry and font size; the profile's position / font / size / color fields are not yet wired into the op.

Text Overflow Prevention

Template placeholders have fixed sizes. To avoid overflow:

• Titles: Max ~60 characters.
• Body bullets: Max 6-7 items per placeholder. For more, split across two slides.
• Bullet text: Keep individual bullets under ~80 characters.
• Emit an OPTIMIZE op after the slide's content to autofit each text box to its shape.

Inject Speaker Notes (real PowerPoint) — SEPARATE PASS

IMPORTANT: Speaker notes MUST be injected as a separate batch pass AFTER all slides exist — never inline during slide creation.

Save the notes map as JSON ({"0": "", "1": "Brief intro.", ...}, 0-based slide indices on the deck as built from the outline), one line per script item. A cue: "BUILD NN" item becomes its own line [BUILD NN], exactly: the build expansion splits a build slide's note at those lines. Then inject via the real PowerPoint app — it writes valid notes OOXML, so the <p:notesMasterIdLst> Keynote patch the old python-pptx pass needed is no longer required (see rules/deck-editing-rules.md):

bash "{speaker_toolkit_root}/skills/presentation-creator/scripts/inject-notes.sh" <uniquely-named deck copy> <out.pptx> notes.json

On an expanded deck, pass the expanded deck copy and notes-expanded.json from the Post-build pass order instead of notes.json.

Run this AFTER slide generation, and BEFORE the final apply-backgrounds.sh pass (the VBA background pass must be the last write). macOS + PowerPoint only.

Keynote compatibility: Real PowerPoint writes the <p:notesMasterIdLst> element natively, so notes-bearing decks open in Keynote with no patch. The old python-pptx pass had to post-process the .pptx to add it — that hack is retired with the python path.

Present to Author

Save and present a generation report with slide count, layouts used, and placeholders needing author content.