Skip to content

slide-deck-author — playbook

Design and outline a slide deck (talk / pitch / reading deck). Outputs slide-by-slide content + design specs that a human or framework (Keynote, Figma Slides, reveal.js / Slidev) can render.

When to use

  • "Make a 10-min talk on X"
  • "Pitch deck for our seed round"
  • "Internal review deck for the new feature"
  • "Conference talk on design tokens"

Inputs (ask if missing)

  1. Type:
  2. Talk deck — presented live, supports speaker
  3. Pitch / sales deck — sent + presented
  4. Reading deck — sent only, read alone (warning: probably should be a doc)
  5. Topic + thesis — what's the deck about? What's the one-line takeaway?
  6. Length — minutes (talk) or slide count (pitch).
  7. Audience — who's watching/reading?
  8. Brand — color, font, logo, voice (or link to brand kit).
  9. Locale — Korean / English / both.

Steps

1. Pick the archetype

If unsure, decision rules from knowledge/patterns/slide-deck-design.md:

| Will it be sent + read alone? | Should be a doc, not a deck | | Will it support a live presenter? | Talk deck (low density) | | Will it be both sent + presented? | Pitch deck (medium density) |

If user insists on a "reading deck": warn that documents render better, then proceed.

2. Draft the outline

Slide-by-slide. Each entry has: - Slide number - Title (the message, not the topic) - Visual (chart / image / mockup / nothing) - Body text or speaker notes

For talk deck (10–25 slides typical):

1. Title slide — "Design Tokens, From First Principles"
2. Hook — the audience problem ("designers and devs say different colors")
3. Personal credibility (~5 sec, optional)
4. Argument 1 — "Tokens unify the language"
   - Chart / before-after image
5. Argument 2 — "Three-tier architecture"
   - Diagram
6. Argument 3 — ...
...
N. Conclusion / call to action

For pitch deck — Sequoia-style 12 slides: 

1. Hero (company + tagline)
2. Problem
3. Solution
4. Why now
5. Product (demo screenshot)
6. Traction (chart)
7. Business model
8. Competition
9. Team
10. Financials / Ask

3. Headline-as-message

For each slide, the title states the takeaway, not the topic.

Topic title (bad) Message title (good)
"Revenue" "Revenue tripled in Q4"
"Architecture" "We chose 3-tier for upgrade safety"
"Onboarding flow" "Onboarding completes in 30 seconds"
"Tokens" "Tokens cut design-dev rework by 40%"

If the audience reads only the slide titles, they get the deck's argument.

4. Choose visuals per slide

For each slide, pick one primary visual: - Chart (per knowledge/patterns/chart-color-encoding.md) - Screenshot of product (with annotations) - Diagram (architecture, workflow, hierarchy) - Photo (for emotion / human moment, sparingly) - Just text (when the message is the visual)

Don't combine multiple visuals on one slide. The visual proves the title.

5. Apply visual / brand

Cite knowledge/patterns/brand-identity.md: - Logo top-left or bottom-right (small). - Brand color for accents (highlights, callouts, chart series). - One typographic family (display + body if pairing). - Background: neutral white or branded solid (no gradient noise).

For Korean-audience decks, cite knowledge/i18n/korean-document-style.md: - Pretendard or similar. - Slightly more text density than Western decks (Korean business convention). - Numbered hierarchical sections.

6. Output spec

Two outputs:

A. Outline markdown (for review)

# Deck: [title]

## Meta
- Type: talk deck
- Length: 10 min (~15 slides)
- Audience: senior designers
- Brand: design-ai

## Outline

### 1. Title
- Visual: company logo + tagline
- Notes: [what speaker says]

### 2. Hook — "Why design tokens matter"
- Visual: side-by-side photo (chaotic palette vs unified)
- Body text: none — image speaks
- Notes: [opening anecdote]

### 3. Argument 1 — "Tokens unify the design-dev handoff"
- Visual: chart of design-dev rework hours, before vs after tokens
- Body text: "Rework hours: -40% after token rollout"
- Notes: [story behind the chart]

...

B. Render-ready format (optional — based on tool)

For Figma Slides / Keynote: outline + visual spec is enough; designer renders.

For reveal.js / Slidev (markdown-based decks): generate the actual .md file:

---
theme: default
title: Design Tokens
---

# Design Tokens
From First Principles

---

# The problem
[image: chaotic-palette.png]

---

# Tokens unify the language
[chart: rework-hours.png]

Verification phase (run before declaring done)

  • Did I pick the right archetype (warn if reading-deck)?
  • Is the title of every slide a message (not a topic)?
  • Each slide has one visual + one message — no mixed?
  • Brand applied (logo, color, type)?
  • Slide count appropriate for length (15-25 for 20-min talk)?
  • If Korean: voice level consistent (~합니다 vs ~해요)?
  • Speaker notes for live talks?
  • No 3D charts / clipart / stock photos of people pointing at laptops?

Source files this skill reads

Done when

  • Outline markdown delivered, slide-by-slide.
  • Each slide title is a message.
  • Each slide has one visual concept + body text.
  • Brand applied consistently.
  • Speaker notes for live decks.
  • (Optional) render-ready format for declarative tools (reveal.js / Slidev).
  • Verification phase passes.