英文 |한국어

导航：工作流程 | 收紧滑套 | 示例 |安装 |许可

一个可重复使用的转弯技能包：

1. 参考幻灯片图像
2. 用户提供的文件
3. 用户提示词/牌组请求

进入一个有纪律的四阶段幻灯片生成工作流程：

1. 从参考幻灯片图像中提取DESIGN.md
2. 在JSON中构建说服性幻灯片计划
3. 用JSON写逐页幻灯提示
4. 从提示JSON顺序生成页面图像

该捆绑包有意按照taste-skill仓库中gpt-taste存在的原因来建模：不是为了增加装饰，而是为了增加更严格的执行、更强的反默认规则以及强制的预检结构，以确保GPT类模型不会跳过步骤或崩溃成通用输出。

为什么这被分为四个技能

单个提示通常以可预测的方式失败：

• 在主题提取之前，它就开始编写幻灯片
• 它将设计分析与牌组策略相结合
• 它会产生没有真正叙事弧线的页面提示
• 它会在身体滑套中失去布局一致性
• 它与参考图像中的可见文字过拟合，而非幻灯片的设计系统

Future Slide Skill

English | 한국어

Navigation: Workflow | Tightened Slide | 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:

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

New command families

Use the newer command families when you want to choose the output mode explicitly.

GPT image slide commands

• $gpt-image-slide → run the full image workflow: design → plan → prompt → render
• $slide-design → only create DESIGN.md
• $gpt-image-slide-plan → only create slide_plan.json
• $gpt-image-slide-prompt → only create slide_prompts.json
• $gpt-image-slide-render → only render page_1.png ... page_N.png

HTML slide commands

• $html-slide → run the full HTML workflow: design → plan → prompt → render
• $slide-design → only create browser-implementable DESIGN.md
• $html-slide-plan → only create slide_plan.json
• $html-slide-prompt → only create html_slide_prompts.json
• $html-slide-render → only build the HTML slide deck

Tightened Slide command