Slidify

Slidify

You generate production-quality PowerPoint presentations using a two-layer architecture:

1. Slide Spec (JSON) — you write and edit this; humans can read and tweak it
2. Code Layer (JS) — generated from the spec; you run it

This separation makes the pipeline AI-friendly: editing means editing JSON, not wrestling with pptxgenjs internals.

Core Philosophy: Images Over Text

1 image > 2 slides of text. Visuals are processed 60,000x faster than text. Every slide with more than 4 bullet points is a candidate for an image replacement.

Before writing any slide, ask: "Can this be a chart, diagram, or image instead of text?"

Use scripts/gen_images.py to generate charts from data. Use scripts/fetch_images.py to find stock images from the web. See references/images.md for the full catalog.

When to Use

• User asks to create, edit, or export a presentation
• User mentions slides, deck, pptx, PowerPoint, or PDF export
• User wants watermarks, logos, brand colors, or footers on slides
• User provides content (paper, report, outline) and wants it as slides
• User wants charts, diagrams, or data visualizations in slides
• User wants stock images or illustrations fetched from the web
• User says "/slidify"

Workflow

Step 1 — Resolve Template

Does the user have a template requirement (university logo, company brand, conference theme)?

Yes → Read references/templates.md, pick or create a template config in assets/templates/. No → Use default template (clean white, no watermark).

Templates live as JSON in assets/templates/. The generator reads them at runtime.

Step 2 — Write Slide Spec

Read references/slide-spec.md for the full schema.

Write slides.json:

{
  "meta": {
    "title": "My Presentation",
    "author": "Your Name",
    "template": "default",
    "output": "output.pptx"
  },
  "slides": [
    { "type": "title", "title": "...", "subtitle": "..." },
    { "type": "content", "title": "...", "layout": "bullets", "items": ["..."], "notes": "Speaker notes here" }
  ]
}

The spec is the single source of truth. All edits go here, never in generated code.

Step 2b — Generate Speaker Notes (MANDATORY)

Every slide MUST have a notes field. Speaker notes appear in the PPTX notes panel below each slide and are essential for presenting — they serve as the presenter's script.

When writing slides.json, add "notes" to every slide object. Generate notes automatically based on the slide content:

What good speaker notes contain:

Notes format — keep them scannable for a presenter:

{
  "type": "content",
  "title": "Key Results",
  "layout": "stats",
  "stats": [
    { "value": "4x", "label": "Token efficiency" },
    { "value": "~2x", "label": "Perplexity gain" }
  ],
  "notes": "OPEN: Let's look at the numbers that matter.\n- 4x token efficiency: this is vs Phi-4 on Hindi — the core contribution.\n- ~2x perplexity: TokenAdapt halves the gap to English-quality models.\nEMPHASIS: Pause on the 4x stat — it's the headline.\nTRANSITION: Now let me show you how we got here..."
}

Slide-type-specific notes:

Step 3 — Generate PPTX

node scripts/gen_pptx.js slides.json
# → output.pptx

The generator reads slides.json + the template config and produces a .pptx.

Step 3b — Generate Images (Optional but Recommended)

If the slide spec has an "images" block, generate charts and fetch stock images:

# Generate charts/diagrams from data
python scripts/gen_images.py images-config.json assets/generated/

# Fetch stock images from the web
python scripts/fetch_images.py fetch-config.json assets/fetched/

Then reference the generated images in the slide spec:

{
  "type": "content",
  "title": "Results",
  "layout": "image-right",
  "image": { "path": "assets/generated/perf_chart.png" },
  "notes": "The chart shows our 4x improvement..."
}

When to generate images:

See references/images.md for all chart types and options.

Step 3c — Add Animations (Optional)

If the user wants animations or transitions, add an "animations" block to slides.json (see references/slide-spec.md for full schema). Then post-process:

node scripts/add_animations.js output.pptx animations.json output-animated.pptx

See references/animations.md for the full effect catalog.

Step 4 — Export (Optional)

bash scripts/export.sh output.pptx pdf    # → PDF
bash scripts/export.sh output.pptx all    # → PDF + PPT + images

Step 5 — QA

# Text content check
python scripts/extract_text.py output.pptx

# Visual check — convert to images
bash scripts/export.sh output.pptx images

Fix issues in slides.json → re-run generator → re-check. Never edit generated code directly.

Slide Types Quick Reference

Content layouts: bullets, two-column, image-right, image-left, cards, stats, timeline, chart, blank

Critical Rules

Editing an Existing Deck

1. Edit slides.json — change content, reorder, add/remove slides
2. Change template? Update meta.template → all slides update automatically
3. Re-run generator — full regeneration from spec (idempotent)

File Map

Composability — Working With Other Skills

See PROTOCOL.md (SIP) at skills root for full interop contract.

Domain Declaration

domain: process
composable: true
yields_to: [craft]

Slidify owns process — the workflow for generating PowerPoint presentations from structured specs. It controls the slide spec format, generation pipeline, and export steps.

When Slidify Leads

• Any request to create, edit, or export a .pptx file
• Any request involving slide decks, presentations, or PowerPoint
• Template creation or modification for presentations

When Slidify Defers

Conflict Signal

⚠️ Process conflict: visual design request overlaps with craft domain. Slidify handles slide structure; craft skill handles aesthetic decisions. Which approach?