Future Slide Skill

English | 한국어

Navigation: Workflow | Examples |
Install | License

A reusable skill bundle for turning:

1. a reference slide image
2. user-provided files
3. a user prompt / deck request

into a disciplined four-stage slide-generation workflow:

1. Extract DESIGN.md from the reference slide image
2. Build a persuasive slide plan in JSON
3. Write page-by-page slide prompts in JSON
4. Generate page images sequentially from the prompt JSON

This bundle is intentionally modeled after the reason gpt-taste exists in the taste-skill repo: not to add more decoration, but to add stricter enforcement, stronger anti-default rules, and mandatory pre-flight structure so GPT-class models do not skip steps or collapse into generic output.

Why this is split into 4 skills

A single prompt often fails in predictable ways:

• it starts writing slides before the theme is extracted
• it mixes design analysis with deck strategy
• it produces page prompts without a real narrative arc
• it loses layout consistency across body slides
• it overfits to the visible text in the reference image instead of the slide design system
• it stops after writing prompts and never actually renders numbered slide outputs

So this bundle separates responsibilities:

• gpt-slide-design → extract a reusable DESIGN.md
• gpt-slide-plan → decide deck logic, ordering, and persuasion
• gpt-slide-prompt → convert that plan into detailed page prompts
• gpt-slide-generate → generate slide images sequentially and save them with page-number filenames

Recommended workflow

Use the skills in this exact sequence:

1) gpt-slide-design

Inputs:

• reference slide image(s)

Output:

• one DESIGN.md focused on presentation design, not slide content

2) gpt-slide-plan

Inputs:

• extracted DESIGN.md
• user files
• user goal / audience / prompt

Output:

• deck plan JSON with slide ordering, narrative flow, and evidence mapping

3) gpt-slide-prompt

Inputs:

• DESIGN.md
• plan JSON
• supporting files if needed

Output:

• slide-by-slide prompt JSON with detailed visual/content instructions

4) gpt-slide-generate

Inputs:

• DESIGN.md
• page-level prompt JSON such as slide_prompts.json

Output:

• final slide images saved sequentially into the workspace, for example:
  • page_1.png
  • page_2.png
  • page_3.png

This step is intentionally separate so generation can:

• inspect each page one by one
• preserve deck consistency across outputs
• save project-bound assets explicitly instead of leaving them in tool cache

Back to top

Example prompts

Below are concrete example prompts based on the actual way these skills were used for the Samsung Biologics / Hana Securities workflow.

gpt-slide-design

Use when you have a reference slide image or reference deck image and want to extract a reusable design system.

Example:

$gpt-slide-design [Image #1]

More explicit example:

$gpt-slide-design
Extract the design theme from this reference slide image.
Focus on official DESIGN.md output with layout placement, header/body/footer flow,
title page / body page / end page flow, icon usage, infographic cards, and diagram behavior.

gpt-slide-plan

Use when you already have DESIGN.md and want to build the storyline and slide sequence from files plus user intent.

Example:

$gpt-slide-plan /Users/tonylee/Downloads/하나증권 _보고서.pdf
Write the analysis slide for 'Samsung Biologics' in Korean based on report pdf file.

Expanded full-deck example:

$gpt-slide-plan /Users/tonylee/Downloads/하나증권 _보고서.pdf
Create a more detailed full deck in Korean from this equity research report.
Keep the structure analytical and report-native.
Plan title page, body pages, end page, and appendix/disclosure flow.

gpt-slide-prompt

Use after planning is complete and you want detailed page-level prompt JSON for slide generation.

Minimal example:

$gpt-slide-prompt

More explicit example:

$gpt-slide-prompt
Use the current DESIGN.md and slide_plan.json.
Generate strict page-by-page prompt JSON with explicit header/body/footer zoning,
table/chart/card hierarchy, icon rules, and anti-generic constraints.

gpt-slide-generate

Use after slide_prompts.json exists and you want actual slide images rendered sequentially.

Minimal example:

$gpt-slide-generate

More explicit example:

$gpt-slide-generate
Based on @slide_prompts.json, create all slide images 1 by 1 and save them
with the page_number naming rule.

Full-deck example:

$gpt-slide-generate
Use DESIGN.md and slide_prompts.json.
Render the full deck sequentially and save:
page_1.png ... page_N.png

Back to top

Example end-to-end usage

Typical sequence:

$gpt-slide-design [reference slide image]
$gpt-slide-plan /path/to/report.pdf Create a full Korean research-summary deck.
$gpt-slide-prompt
$gpt-slide-generate

What this bundle optimizes for

• theme extraction first, content second
• strong cross-slide consistency
• persuasive story flow
• explicit evidence-to-slide mapping
• reusable layout families
• body-slide discipline
• title / body / end-page flow discipline
• explicit header / body / footer zoning
• icon / infographic / table / chart role clarity
• sequential image generation with stable numbering
• no hallucinated design rules when the reference image is ambiguous

Output artifacts

This bundle includes:

• skills/gpt-slide-design/SKILL.md
• skills/gpt-slide-plan/SKILL.md
• skills/gpt-slide-prompt/SKILL.md
• skills/gpt-slide-generate/SKILL.md
• templates/DESIGN_TEMPLATE.md

Current skill responsibilities

gpt-slide-design

• extracts design theme, placement rules, header/body/footer flow, title/body/end-page behavior
• captures icon usage, infographic card logic, table/chart treatment, and diagram behavior

gpt-slide-plan

• builds the storyline and page-family rhythm
• decides where tables, charts, icon-led modules, or comparison exhibits belong
• plans split topics across multiple pages when needed

gpt-slide-prompt

• converts the plan into strict per-page prompt JSON
• makes layout family, zoning, and anti-generic rules explicit for every page

gpt-slide-generate

• reads DESIGN.md and prompt JSON
• renders slide images one by one with Codex native image generation
• saves final outputs into the project using page-number naming

Recommended generated artifact set

For a full run, the typical artifact chain is:

1. DESIGN.md
2. slide_plan.json
3. slide_prompts.json
4. page_1.png ... page_N.png

Back to top

Installation

Install with npx skills

Use the Skills CLI from a terminal with Node.js 18+:

npx skills add jyoung105/future-slide

You can also use the repository URL:

npx skills add https://github.com/jyoung105/future-slide.git

Restart Codex after installation so the new skills are discovered.

Manual install into .codex

Download or clone this repository, then copy the skill folders into your
Codex skills directory:

mkdir -p ~/.codex/skills
cp -R skills/gpt-slide-design ~/.codex/skills/
cp -R skills/gpt-slide-plan ~/.codex/skills/
cp -R skills/gpt-slide-prompt ~/.codex/skills/
cp -R skills/gpt-slide-generate ~/.codex/skills/

For project-local installation, copy the same folders into:

.codex/skills/

Codex discovers skills from folders that contain SKILL.md, so each copied
folder must keep its SKILL.md at the folder root.

Back to top

License

Future Slide Skill is released under the Apache License 2.0.

Back to top