GordenSuperPPTSkill

GordenSuperPPTSkill — End-to-end PPT full-process orchestration

Run the two sub-skills in sequence:

内容 ─▶【阶段1 GordenImagePPTGen】图片型 PPT（每页 .png + 图片版 .pptx）
       ─▶【阶段2 GordenImage2PPTX】对每页图片做 背景图+骨架图+元素图标+文本 还原 ─▶ 可编辑 .pptx

This skill itself does not reinvent the wheel—itexecutes both sub-skills completely in sequence. Both sub-skills need to be co-located with this skill (same repository/same skills directory):

Stage 1:../GordenImagePPTGen/SKILL.md(Function A: generate image-based PPT)
Stage 2:../GordenImage2PPTX/SKILL.md(Function B: image → editable PPTX)

For the end-to-end runbook, see references/pipeline.md。

Stage 1 image generation hard gate

In Stage 1, "generate images" can only be interpreted as:actually calling Codex's built-in imagegen image model to generate each finished slide image。

It is forbidden to replace Stage 1 image output using the following methods:

Python/PIL, SVG, HTML, Canvas, matplotlib, PowerPoint shapes, screenshot rendering.
Using code to draw an entire slide image, then treating it as an "image-based PPT".
First making pages with editable PPTX / native shapes, then exporting them as PNG to serve as the Stage 1 result.
Using code to add text, overlay text, or modify text on an already generated image.

The Codex runtime must satisfy:

Each page calls imagegen one or more times, and finally one model-generated image is selected.
The generated source image must come from $CODEX_HOME/generated_images/<thread-id>/...png。
The source image may only becopiedto this task's slides/NN-*.png; do not delete the original image in the default generation directory.
Must write imagegen-manifest.json, recording page by page generated_source and copied_to。

If imagegen is unavailable, you must stop and explain the blocking reason; you cannot fall back to code drawing.

When to use which

User intent	Which skill to use
"Make a PPT / generate an image-version PPT / AI-generated-image slides"	Use only GordenImagePPTGen
"Convert these PPT images/screenshots into an editable PPTX / cut out icons / extract text"	Use only GordenImage2PPTX
Not specified by name, only a topic/content is given and a usable finished product is wanted; or "both viewable and editable" is wanted	Use this skill(A→B chained)

Userexplicitlynames a certain function → directly use the corresponding sub-skill; no need to go through Super.
When not named, default to Super: first complete image generation, then restore page by page to editable, and finally deliver all artifacts at once.

Orchestration flow (check off item by item)

== 阶段 1：GordenImagePPTGen（完整跑功能 A）==
- [ ] 读 ../GordenImagePPTGen/SKILL.md，按 A1–A5 执行
- [ ] 每页实际调用 imagegen 生成成品图
- [ ] 产出：outline.json、prompts/NN-*.md、imagegen-manifest.json、slides/NN-*.png、图片型 .pptx
- [ ] 铁律：每页图必须含【全部真实文字 verbatim】，绝不出占位符/无字模板图
- [ ] 门禁：没有 imagegen-manifest.json 或任一页缺 generated_source，则阶段 1 失败，不能进入阶段 2

== 阶段 2：GordenImage2PPTX（对阶段1每张图完整跑功能 B）==
- [ ] 读 ../GordenImage2PPTX/SKILL.md，对每页图片按 B1–B9 执行
- [ ] 强制四层：背景图 → 骨架图(绿幕抠图) → 元素图标(绿幕抠图) → 文本(OCR)，缺一不可
- [ ] 背景图、骨架图、元素图标图必须由 imagegen 提取式生成
- [ ] 禁止用 PPT 色块、原生 shapes、代码绘图、裁原图局部替代背景/骨架/图标图片层
- [ ] 产出：每页 imagegen-assets-manifest.json + background/frame/icons/layout.json + 合成的可编辑 .pptx

== 交付 ==
- [ ] 一次性交付：① 所有 PPT 图片 ② 每页背景图 ③ 每页骨架图 ④ 每页图标/装饰 ⑤ 每页文本数据(layout.json) ⑥ 图片型 .pptx ⑦ 可编辑 .pptx

Key constraints (must be observed in both stages)

🔴 Stage 1Must call imagegen to generate images, never replace with code drawing, and never output placeholder images; Stage 2the background/skeleton/icon image layers must also call imagegen to generate, never skip skeleton images and element icons(four layers mandatory). These two are the root causes of the previous version's degradation, so they must be safeguarded.
Unified aspect ratio: A and B use the same aspect ratio throughout (default 16:9; if the user wants 3:2, use 3:2 for the whole set).
Unified color scheme, no repeated layout on each page(Stage 1);cutouts preserve color and lines, icon sizes match the original image(Stage 2).
No fabricated data; language follows the user; keep the footer clean; avoid pure green (it conflicts with green-screen cutouts).

Output directory structure (recommended)

<topic-slug>/
├── outline.json / imagegen-manifest.json / prompts/ / slides/   # 阶段1
├── out/<topic>-image-deck.pptx              # 阶段1 图片型成品
└── editable/                                # 阶段2
    ├── NN/{background.png, frame.png, icons/*.png, layout.json}
    ├── deck.json
    ├── preview/slide_*.png
    └── <topic>-editable.pptx                # 阶段2 可编辑成品

Each sub-skill comes with its own scripts and references; this skill is only responsible for orchestration and delivery. If the two sub-skills are not in the same location, install them first, or call them separately.