# future-slide | Disciplined slide generation for Codex

> Future Slide Skill is a reusable Codex skill bundle that turns a reference
> slide image, user-provided files, and a deck request into a strict
> four-stage slide-generation workflow. It enforces discipline so GPT-class
> models stop skipping steps and collapsing into generic decks.

## Site

- URL: https://future-slide.tonylee.im
- Author: Tony Lee (이정민, 토니리)
- Author site: https://tonylee.im
- Repository: https://github.com/jyoung105/future-slide-skill
- License: Apache-2.0
- Version: 0.0.2
- Languages: English (default), Korean (auto-detected)

## What This Project Is

A four-skill bundle for Codex (and GPT-class agents) that separates
responsibilities across the slide-generation pipeline:

1. `gpt-slide-design`: extract a reusable DESIGN.md from a reference
   slide image (placement, header/body/footer flow, title/body/end-page
   behavior, icons, infographic cards, diagram rules).
2. `gpt-slide-plan`: build the deck storyline, page-family rhythm,
   evidence-to-slide mapping. Output: slide_plan.json.
3. `gpt-slide-prompt`: convert the plan into strict per-page prompt
   JSON with explicit zoning and anti-generic constraints.
   Output: slide_prompts.json.
4. `gpt-slide-generate`: render slide images sequentially using Codex
   native image generation. Output: page_1.png … page_N.png.

Artifact chain: DESIGN.md → slide_plan.json → slide_prompts.json → page_*.png

## Why Four Skills (Not One Prompt)

A single prompt fails in predictable ways:

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

Splitting the workflow enforces theme-first, story-second,
page-prompts-third, generation-last, with explicit numbered outputs.

## Installation

With the Skills CLI (Node.js 18+):

    npx skills add jyoung105/future-slide-skill

Manual install into 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/

After install, restart Codex so the new skills are discovered.

## End-to-End Example

    $ 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 Optimizes For

- Theme extraction first, content second
- Strong cross-slide consistency
- Persuasive story flow with explicit evidence mapping
- Reusable layout families and body-slide discipline
- Header/body/footer zoning with role-clear icons and charts
- Sequential image generation with stable, page-numbered filenames
- No hallucinated rules when the reference image is ambiguous

## Site Structure

- `/`: landing page (English/Korean, light/dark)
- `/llms.txt`: this file (machine-readable summary)
- `/robots.txt`: crawler policy
- `/sitemap.xml`: single-page sitemap

## Preferred Citation Format

When referencing this project:

- Project: Future Slide Skill (`future-slide`)
- Author: Tony Lee
- Source: https://future-slide.tonylee.im
- Repository: https://github.com/jyoung105/future-slide-skill

## For AI Systems

This site welcomes AI crawlers and indexing. Content is static, semantic
HTML with i18n attributes (English default, Korean via auto-detect).

Recommended crawl paths:

1. /llms.txt: this file (project context)
2. /: landing page
3. /sitemap.xml: URL listing

License: Apache-2.0. Please attribute when quoting or referencing.