Slides

You are generating a presentation. Follow these steps in order.

Step 1 — Determine inputs

Extract from the user's message:

• topic: what the presentation is about (required — ask if missing)
• format: html (reveal.js, default) or pptx (PowerPoint)
• output_dir: where to save files (default: slides/examples/ relative to project root)
• css: custom CSS path for HTML only (default: slides/templates/mongodb-revealjs.css if that file exists, otherwise omit)
• theme: reveal.js base theme for HTML only (default: black)

Step 2 — Write the markdown source

Save <output_dir>/<Topic>.md. Structure:

• # Title — one H1 (title slide)
• ## Slide Title — each H2 is one slide (pandoc --slide-level=2)
• 4–6 bullets or one short code block per slide — no walls of text
• For a project pitch, use this arc: Problem → Solution → How It Works → Architecture → What's Next → Questions
• Adapt the arc to the actual content. Cut slides that don't serve the story.
• Note for pptx: avoid {width=…} image attributes and reveal.js-specific fragment syntax — they are ignored or may cause warnings in pptx output.

Step 3 — Build with pandoc

HTML (reveal.js)

pandoc "<input.md>" -t revealjs -s \
  -o "<output.html>" \
  --slide-level=2 \
  -V theme=<theme> \
  [--css "<css_file>"]

Omit --css if no CSS file is available.

PowerPoint (.pptx)

pandoc "<input.md>" -t pptx \
  -o "<output.pptx>" \
  --slide-level=2 \
  [--reference-doc="slides/templates/reference.pptx"]

Omit --reference-doc if the template does not exist. To apply team branding, replace slides/templates/reference.pptx with a branded file — pandoc copies its theme (colors, fonts, layouts) into the output.

Or use the helper script (handles both formats):

./slides/build.sh "examples/<Topic>.md"           # HTML
./slides/build.sh "examples/<Topic>.md" --pptx    # PowerPoint

If pandoc is missing, tell the user to run: brew install pandoc (or install.sh in the project root).

Step 4 — Ask about narration

Before building, ask the user: "Should I also generate a narration script you can play with say?"

If yes (or they already said --narrate / "with narration"):

1. Read the markdown source you just wrote.
2. For each slide (H1 title slide, then each ## section), produce clean speakable text:
   • Remove code fences and inline code
   • Remove heading markers, bold/italic, bullet prefixes, HTML comments, image/link syntax
   • Keep the natural sentence flow — rewrite fragments into full phrases if needed
3. Write the result to <output_dir>/<Topic>-narration.txt, slides joined by [[slnc 1500]].
4. Tell the user to run:

   say -f "<output_dir>/<Topic>-narration.txt"
   # or with voice/rate:
   say -v Samantha -r 175 -f "<output_dir>/<Topic>-narration.txt"
   

Step 5 — Report

Tell the user:

• Markdown source path (edit this to update content)
• Output path and how to open it
• Exact pandoc command (or build.sh invocation) to rebuild after edits
• If narration was generated: the narration file path and the say command to run it
• For pptx: remind them that slides/templates/reference.pptx is the branding hook — replace it with a branded template to apply team styles