luobo-ppt skill
A skill that builds a deck from a topic or article, or restyles an old one from a reference image, keeping the text and editable slides
- 452 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill runs Python scripts and installs missing packages through pip on first run
- Deck content and your pictures go to an outside service if you connect image generation
- Photo search in open libraries uses keys kept in environment variables
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/luobo-pptDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill runs Python scripts and installs missing packages through pip on first run
- Deck content and your pictures go to an outside service if you connect image generation
- Photo search in open libraries uses keys kept in environment variables
This is third-party code. Review the repository files before installing.
What it does
luobo-ppt guides an agent down two roads that share one design engine. The first builds a deck from scratch: you give a topic, an article, a speech text or an outline, and the skill writes conclusion-style headings with speaker notes, first shows a titles-only ghost draft for your confirmation, then picks layouts, icons and a style. The second takes a plain deck and a reference image and carries the text over word for word into a new palette, layouts, icons and illustrations. The result is always a plain .pptx file. Text sits in text boxes, tables are real tables, charts are native PowerPoint charts with editable data, and the files open in PowerPoint, WPS and Keynote. Layouts cover a timeline, funnel, 2x2 matrix, SWOT, PEST, Porter's five forces and a business canvas, with separate layouts for screenshots with numbered markers. Equations made with the equation editor are converted to plain text so they do not vanish in other programs. Before handing over, scripts check for lost text, overflow, overlap with the footer, contrast and the number of accent colors, and the agent scores itself on five criteria. You can supply a company template in .pptx or .potx, and slides are built on its master, or pick one of fourteen ready styles.
Who it is for. For people who prepare talks, courses, reports and tool walkthroughs and want slides that can later be edited by hand in an ordinary editor.
Good fit when
- You need to turn an article or a speech text into a deck with conclusion headings and speaker notes
- You have a plain deck that must be restyled from a reference image without changing a word
- The deck has to be built on a company .pptx or .potx template with a logo and footer on every slide
- The deck is full of formulas, as in a school course, and they must stay visible in any program
- The deck is shown in a meeting room through a projector and you need a larger-font mode
Not a fit when
- You want a picture-slide deck where each page is generated as one image: the skill builds editable slides and makes no slide images
- You need to rewrite the content and structure of an old deck: in restyle mode the text stays unchanged, so use the generation route
- You have no Python 3.8 or newer: the build and check scripts run on it
- You want elaborate animation with complex effects: the skill sticks to a gentle page fade and step-by-step appearance of items
Example request
Turn this article into a 12-slide deck with speaker notes in the ink-magazine styleLimitations
The skill's documentation and triggers are written in Chinese, and a font with Chinese characters is listed among the required dependencies. It requires Python 3.8 or newer and the python-pptx, Pillow, numpy and lxml libraries. If LibreOffice is not installed, the preview is an approximate render from layout data, and the report says so.
Pictures and illustrations are drawn by program by default. To connect image generation you need a key for OpenAI, Gemini, Seedream, HiAPI or a compatible service. The skill asks you to keep the key in environment variables and not write it to files. Photos from open libraries are taken only under licenses that allow commercial use, and those that require attribution go into a separate list. A full pass over 12 pages takes 15 to 25 minutes, and quick mode removes only the questions to you, with the checks kept.
How to disable. Delete the luobo-ppt folder from the agent's skills directory, for example `rm -rf ~/.claude/skills/luobo-ppt`, and open a new session. Working files sit in the project's `ppt_work` folder, which you delete separately.
FAQ
How do I install luobo-ppt in Claude Code?
Tell the agent to install this skill and give it the repository address, or run `git clone https://github.com/zhouwei713/luobo-ppt.git ~/.claude/skills/luobo-ppt`. With the skills installer, `npx skills add zhouwei713/luobo-ppt` works. Open a new session afterwards.
How does the skill make sure text is not lost during restyling?
A script stores every source paragraph and compares them page by page after the build. If even one line vanishes, it reports the slide number and the phrase. Charts, notes and sources are checked separately, symbols included.
Can I run it without LibreOffice?
Yes. LibreOffice gives an exact preview, and without it the render_lite.py script draws approximate pages from layout data: about 10 seconds for 12 slides. Layout, colors and overflow are still checked.
Which styles exist without a reference image?
Fourteen ready ones: mint fresh, business blue, warm orange, dark tech, lavender, paper notes, black gold, geometric blocks, cobalt type, ink magazine, Swiss grid and others. The skill shows three samples on your own slides to choose from.
Can slides be built on a company template?
Yes. Pass a .pptx or .potx template: the logo, color bars and footer on the master stay, colors and fonts come from the template theme, and content is placed only in free areas. Save old .ppt and .pot formats as .pptx first.
Do I need a key to generate pictures?
No. Without a key the program draws backgrounds, decorations and illustrations. A key connects a model for richer pictures, and if one picture fails the skill falls back to the program-drawn one.
Discussion
No questions or reviews yet.
Related
SkillsA skill that builds presentations from HTML slides with rendering, per-slide visual review, and conversion to an editable PPTX
A skill that builds a deck through dialog and previews: structure and style first, then page images generated and packed into PPTX
A skill for talks: outlines a deck around the audience's train of thought, picks images and video, and checks finished slides before you present
A skill that builds decks with gpt-image-2: each slide is drawn as a full layout and packed into .pptx, optionally mimicking a reference template
A skill that builds editable PPTX decks from PDF, DOCX, URLs and Markdown with native shapes, charts and animations
A deck from a single sentence: seven stages, a sub-agent per slide, screenshot-based QA and PPTX export.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| luobo-ppt (this card) | no | 452 | |
| PPT Master | no | 73,972 | builds editable decks from PDF, DOCX, URLs and Markdown with native shapes |
| figedit | no | 193 | turns raster figures into editable vector slides with text recognition |
| gpt-image2-ppt-skills | no | 1,426 | draws each slide whole with an image model, so slides come out as pictures |
README in short
The Chinese description opens with an animation of a plain slide turning into a styled one and three key capabilities. Result examples follow, then a list of typical flaws of AI decks, a data-flow diagram and four ways to install. A style table, dependencies, sample requests and a detailed changelog up to 2.9.0 close it.
README badge
Original SKILL.md
--- name: luobo-ppt description: 做 PPT 和美化 PPT 的一站式 skill，输出可编辑的原生 .pptx。从零生成时，给一个题目、一篇文章或口播稿、一份大纲，产出结论式标题、带口播备注、风格统一的 PPT；按参考图美化时，给一份朴素的 .pptx 和一张参考图，文字一字不改地换上统一的配色、母版、版式、图标和插画。当用户说"做个PPT""把这篇文章做成PPT""生成口播PPT""美化PPT""照这张图的风格改PPT""套个风格""用公司模板做""make slides""beautify my deck"，或丢来 pptx、potx 模板、文章、大纲时，务必使用此skill。生图模型可配置；没配置时先问用户，可以现场接入模型、搜可商用的公开图库实拍图、用自己的素材，或用程序化插画。口播和讲课的稿子自带逐条出现的动画。 --- # 萝卜 PPT（luobo-ppt） 两条路线共用一套设计引擎。生成是"写出内容再设计"，美化是"抽出原稿内容再设计"，从版式规划开始完全一样。 三份中间文件串起全流程，脚本负责确定性的活，Claude 负责理解内容和审美判断。 | 文件 | 内容 | 美化时 | 生成时 | |---|---|---|---| | `deck.json` | 每页的页型、标题、要点、数据、备注 | 从原稿抽取，文字一字不改 | 按 `references/generate.md` 写出来 | | `style.json` | 色板、字体、圆角、装饰、插画风格 | 从参考图抽取 | 参考图抽取，或从预设库挑 | | `plan.json` | 每页版式、图标、插画和提示词 | `plan_deck.py` 起草，Claude 审改 | 同左 | 下文 `$S` 指本 skill 的 scripts 目录，`$W` 指本次的工作目录（如 `./ppt_work`）。Windows 上把 `python3` 换成 `python`。 ## 开工前 1. **判断路线** - 用户给了 .pptx，要换个好看的样子 → 路线 A 美化 - 给的是题目、文章、口播稿、大纲或其他素材，要做成 PPT → 路线 B 生成 - 给了 .pptx 但要重写内容、调整结构 → 路线 B，把 pptx 当素材读 - 另外给了企业 VI 手册、品牌规范，或者说"按公司 VI 做""用我们的品牌色"→ 两条路线都先做"品牌套件"，风格从手册来，不再读参考图、不出风格样张 - 另外给了公司的 PPT 模板（.pptx 或 .potx），或者说"用公司模板做""套我们的模板"→ 两条路线都先做"公司模板"，在模板上出稿，不再读参考图、不出风格样张。模板和 VI 手册同时给时以模板为准，手册只拿来核对颜色 2. **环境自检**，之前通过的话一秒内结束，否则自动补装依赖。退出码非 0 就把建议告诉用户并停下；只缺推荐组件（如 LibreOffice）可以继续，pipeline 会改用 `render_lite.py` 按排版数据直接画近似预览，排版、配色、溢出照样能看，交付时说明预览是近似图。想跳过 LibreOffice 提速也可以加 `--lite`。 ```bash python3 $S/setup_check.py --quick --install ``` 3. **生图状态** `python3 $S/make_assets.py $W --status`，用一句话告诉用户接了哪家模型。没有模型时先别默认走程序化插画，等版式规划出来、确实需要配图时按下面的"配图来源"问用户。配置方法见 `references/imagegen_config.md`。 4. **一次提问**问清路线需要的选择，其余用默认值。用户说"直接做""不用问"或者不在场时，全部用默认值，交付时列出做过的选择。 5. **快速模式**：用户说"快速模式""别问了""赶时间"时，跳过风格卡、大纲、设计规格的确认，五维自评只修一轮"要修"，"顺手改"写进交付说明不动手。快速模式只省掉等用户确认的环节，逐字核对、图表和小字核对、全部 QA 检查照常跑，ERROR 没清零就不交付。12 页左右的稿子，正常 15 到 25 分钟，快速模式 10 分钟上下；时间大头在转写文字和看图，脚本本身只要一两分钟。 - 美化：力度 light / standard / full（默认 full），是否允许精修文字（默认不允许） - 生成：用途、受众、时长、有无参考图、要不要口播备注，默认值见 `references/generate.md` ## 路线 A 美化 1. **解析原稿** `python3 $S/extract_deck.py 原稿.pptx $W`，然后看 `source_png/sheet.png`。脚本会修复 LibreOffice 打不开的 zip，也能读出藏在 descr 属性里的口播备注。 2. **读参考图** `python3 $S/extract_style.py 参考图.png $W`，再亲自看图，按 `references/style_extraction.md` 写 `$W/style.json`，然后 `python3 $S/style_card.py $W` 出风格卡。用户在场就发风格卡确认。 3. **整理 deck.json**，按 `references/deck_schema.md` 从 `deck.draft.json` 转写。文字逐字复制，局部加粗写成 `**片段**`；图片里的文字重建成 `diagram` 等原生结构；`notes` 原样带上；写好 `meta.anchor` 视觉锚点。原稿页面上的小标签（"01 · 概览"这类章节胶带）写进该页的 `kicker`，条目自带的编号（01、02）写进条目的 `num`，这样换任何列表符号都不会丢字。原稿里用公式编辑器打的公式，解析时会自动转成普通文字公式（如 V′(x) = (a−2x)(a−6x)），照常逐字转写；公式多的讲解页用 `explain` 页型。SVG 图会顺手转一份 PNG（raw.json 里的 png 字段），版式里用 PNG。logo、"AI 生成"标识这类小图写进 `stamps` 原样保留。原稿里的图表在 `deck.draft.json` 每页的 `charts` 里已经整理成现成的图表写法，用 `chart` 页型照抄进去，`number_format`、`title`、`x_title`、`y_title` 一个都不能丢，丢了图上的单位和坐标轴说明就跟着没了。一页有两三个图表时整组抄进 `charts`，散点图带着每组的 `x` 原样抄。原稿表格在 `tables` 里，抄进 `table` 页型。页面底部带 * ※ 的注释、来源，role 是 footnote，写进该页的 `footnote` 和 `source`，两个可以同时写。原稿里的软件截图（`media/` 下的图片）用 `screenshot` 或 `screenshots` 页型，截图会放进统一的窗口外框，讲解点可以用 `at` 打在截图上。 4. 接着走下面的"版式、出图与检查"。