chore: import upstream snapshot with attribution

.gitignore

@@ -0,0 +1,12 @@
+.DS_Store
+__pycache__/
+*.py[cod]
+
+# Local generation outputs
+generated/
+outputs/
+tmp/
+
+# Editor/project noise
+.idea/
+.vscode/

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Ian
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

NOTICE.md

@@ -0,0 +1,12 @@
+# Notice
+
+Ian Handdrawn PPT is an AI-agent skill created by Ian.
+
+The bundled reference image in `ian-handdrawn-ppt/assets/reference-handdrawn-article-illustration-style.png` was generated by Ian and is included as a style anchor for this skill.
+
+When redistributing or adapting this repository, please keep the `Ian Handdrawn PPT` name or provide attribution to Ian in your derived documentation.
+
+Author links:
+
+- GitHub: <https://github.com/helloianneo>
+- Website: <https://ianneo.xyz>

README.md

@@ -0,0 +1,258 @@
+# Ian Handdrawn PPT
+
+> 把文章、课程笔记、提纲或已有材料，变成一套中文手绘技术解释风格的 PPT-style 页面图。
+>
+> 21:9 封面 | 16:9 正文配图 | PNG 输出 | 适合文章配图、课程课件、知识卡片和技术概念解释
+
+---
+
+## 这个仓库是什么
+
+Ian Handdrawn PPT 是一个 Codex Skill，用来指导 AI Agent 把内容整理成**中文手绘技术解释图**。
+
+它不是传统 PPT 模板，也不是可编辑 PPTX 生成器。它的目标是产出可以直接用于文章、课程、分享稿里的**整页 PNG 视觉图**：
+
+- 先理解材料，提炼叙事结构
+- 再把每一页映射到合适的语义版式
+- 统一锁定手绘视觉 DNA
+- 每页生成一张完整 raster image
+- 多页时生成 contact sheet 方便快速检查
+
+一句话：**让 AI 不只是“做几页 PPT”，而是先想清楚内容，再生成一组有统一风格的中文技术解释图。**
+
+推荐生图模型：**最好使用 ChatGPT Image 2.0**。如果当前环境没有这个模型，就使用可用的最高质量图像模型，并尽量让整套图保持同一个模型、同一套 style lock。
+
+---
+
+## 适合谁用
+
+特别适合：
+
+- 写中文技术文章，需要封面图和正文配图的人
+- 做课程、训练营、工作坊，需要解释复杂概念的人
+- 想把长文、提纲、讲稿变成视觉化页面的人
+- 用 Codex 做内容生产，希望减少“模板 PPT 味”的人
+- 想让 AI 先规划页面叙事，再生成图像的人
+
+不适合：
+
+- 想要可编辑 PPTX 源文件的人
+- 想要自动导出 PDF / PPTX 文件的人
+- 想要复杂动效、交互式网页或矢量图的人
+- 想把大量正文塞进一张图里的人
+
+---
+
+## 它会产出什么
+
+默认输出：
+
+- 21:9 文章封面图
+- 16:9 正文插图 / 标准页面图
+- 多页 contact sheet
+- 简短 slide blueprint summary
+
+默认不输出：
+
+- 可编辑 PPTX
+- image-based PPTX
+- PDF
+- Keynote 文件
+- HTML / SVG / canvas 版本
+
+---
+
+## 视觉风格
+
+这个 skill 默认使用 Ian 的中文手绘技术解释风格：
+
+- 近白纸底，不发黄
+- 无整页边框
+- 细手绘线条和轻微铅笔排线
+- 淡蓝、鼠尾草绿、浅桃、淡紫标记
+- 中央图小而精，保留大量空白
+- 标题克制，不做大字海报
+- 中文文字短、少、可检查
+- 人物很少，最多一个小读者/工程师角色
+
+参考图在：
+
+```text
+ian-handdrawn-ppt/assets/reference-handdrawn-article-illustration-style.png
+```
+
+---
+
+## 示例效果
+
+下面是一组用 Ian Handdrawn PPT 生成的示例图：1 张超宽封面 + 3 张正文解释页。
+
+### 能自动化不等于该自动化
+
+![能自动化不等于该自动化](examples/images/cover-automation-boundary.png)
+
+### 阅读其实是两件事
+
+![阅读其实是两件事](examples/images/page-01-reading-two-things.png)
+
+### 自动化该放哪里
+
+![自动化该放哪里](examples/images/page-02-where-to-automate.png)
+
+### 三问判断法
+
+![三问判断法](examples/images/page-03-three-question-method.png)
+
+---
+
+## 安装
+
+克隆仓库：
+
+```bash
+git clone https://github.com/helloianneo/ian-handdrawn-ppt.git
+cd ian-handdrawn-ppt
+```
+
+复制 skill 到 Codex skills 目录：
+
+```bash
+mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
+cp -R ./ian-handdrawn-ppt "${CODEX_HOME:-$HOME/.codex}/skills/"
+```
+
+安装后，在 Codex 里使用：
+
+```text
+Use $ian-handdrawn-ppt 把这篇文章做成 1 张封面图 + 3 张正文配图。
+```
+
+---
+
+## 怎么用
+
+### 文章封面 + 正文配图
+
+```text
+Use $ian-handdrawn-ppt 把下面这篇文章做成 1 张 21:9 中文封面图 + 3 张 16:9 正文配图。
+风格保持中文手绘技术解释图，文字尽量短，每张图只表达一个观点。
+
+<粘贴文章>
+```
+
+### 课程课件页
+
+```text
+Use $ian-handdrawn-ppt 把这份课程大纲整理成 8 页中文手绘技术课件图。
+面向有基础的新手，每页一个核心概念，输出 slide-by-slide blueprint，然后生成最终 PNG 页面图。
+
+<粘贴课程大纲>
+```
+
+### 只规划，不生图
+
+```text
+Use $ian-handdrawn-ppt 先不要生图。
+请把这篇内容规划成一套 10 页左右的中文手绘技术 PPT-style image deck。
+每页给出标题、主旨、版式 archetype、可见文字和图像 brief。
+
+<粘贴素材>
+```
+
+更多示例见 [examples/prompts.md](examples/prompts.md)。
+
+---
+
+## 工作流程
+
+这个 skill 的流程是：
+
+1. 读取材料：文章、Markdown、PDF、DOCX、PPTX、课程大纲、讲稿或粗略想法
+2. 做 intake：判断主题、受众、场景、目标、核心论点和素材充分度
+3. 规划叙事：选择教学、说服、报告、产品解释或知识卡片结构
+4. 选择页面 archetype：封面隐喻、左右对比、流程、循环、分类、矩阵、总结等
+5. 锁定视觉 DNA：统一纸底、标题、页码、线条、色彩、图形尺度
+6. 每页单独写图像 prompt：包含页面角色、主旨、构图和 `Required text only`
+7. 生图并检查：中文文字、画幅、风格一致性、是否漂黄、是否过度模板化
+8. 多页生成 contact sheet，最后交付 PNG 页面图
+
+---
+
+## 目录结构
+
+```text
+.
+├── README.md
+├── LICENSE
+├── NOTICE.md
+├── examples/
+│   ├── images/
+│   │   ├── cover-automation-boundary.png
+│   │   ├── page-01-reading-two-things.png
+│   │   ├── page-02-where-to-automate.png
+│   │   └── page-03-three-question-method.png
+│   └── prompts.md
+└── ian-handdrawn-ppt/
+    ├── SKILL.md
+    ├── assets/
+    │   ├── reference-handdrawn-article-illustration-style.png
+    │   └── theme-tokens.json
+    └── references/
+        ├── intake.md
+        ├── narrative-planning.md
+        ├── output-quality.md
+        ├── prompt-patterns.md
+        ├── slide-archetypes.md
+        └── visual-dna-v6.md
+```
+
+真正需要安装到 Codex 的是子目录：
+
+```text
+ian-handdrawn-ppt/
+```
+
+根目录的 README、LICENSE、NOTICE 和 examples 是 GitHub 分享文档。
+
+---
+
+## 注意事项
+
+- 图片里的中文文字越短越稳定。
+- 最好每个页面多生成几次，再从中挑选中文字、构图、语义关系都最稳的一张。
+- AI 图像模型可能出现错字、幻觉标签、物体关系错误、风格漂移，不要默认第一张就是最终稿。
+- 如果中文渲染失败，应先减少字数并重生。
+- 如果画面方向已经正确但文字仍错，可以保留图像方向，再用确定性后处理叠准确中文。
+- PPTX/PDF 可以作为输入素材读取，但不是这个 skill 的输出格式。
+- 这个 skill 的重点是“内容语义 → 页面叙事 → 手绘解释图”，不是模板套版。
+
+---
+
+## 相关资料
+
+如果你对 AI 工作流、内容创作和个人知识库感兴趣，也可以看这些：
+
+- [Awesome Claude Code Skills](https://github.com/helloianneo/awesome-claude-code-skills) — Claude Code Skills / Agents / Plugins 精选合集
+- [Obsidian + Claude AI Second Brain](https://github.com/helloianneo/obsidian-ai-second-brain) — Obsidian + Claude AI 个人知识库搭建指南
+- [Claude Code Handbook](https://github.com/helloianneo/claude-code-handbook) — Claude Code 高阶使用指南
+
+---
+
+## 关于作者
+
+**Ian (伊恩)** — 产品设计师 / 一人公司实践者 / AI Builder
+
+用 AI 团队打造一人公司。
+
+- X/Twitter: [@ianneo_ai](https://x.com/ianneo_ai)
+- 网站: [ianneo.xyz](https://ianneo.xyz)
+- 微信: 17855813746
+- 邮箱: hello.neoc@gmail.com
+
+> **🚀 正在开启：内容创作丨AI 工作流围观营，欢迎来玩！** 加微信了解详情。
+
+---
+
+## License
+
+MIT License. See [LICENSE](LICENSE).

README.wehub.md

@@ -0,0 +1,7 @@
+# WeHub 来源说明
+
+- 原始项目：`helloianneo/ian-handdrawn-ppt`
+- 原始仓库：https://github.com/helloianneo/ian-handdrawn-ppt
+- 导入方式：上游默认分支的最新快照
+- 原作者、版权和许可证信息以原始仓库及本仓库 LICENSE 为准
+- 本文件仅用于记录来源，不代表 WeHub 是原项目作者

examples/prompts.md

@@ -0,0 +1,45 @@
+# Prompt Examples
+
+下面这些 prompt 可以直接复制到 Codex 里使用。
+
+## 文章封面 + 正文配图
+
+```text
+Use $ian-handdrawn-ppt 把下面这篇文章做成 1 张 21:9 中文封面图 + 3 张 16:9 正文配图。
+风格保持中文手绘技术解释图，文字尽量短，每张图只表达一个观点。
+
+<粘贴文章>
+```
+
+## 课程课件页
+
+```text
+Use $ian-handdrawn-ppt 把这份课程大纲整理成 8 页中文手绘技术课件图。
+面向有基础的新手，每页一个核心概念，输出 slide-by-slide blueprint，然后生成最终 PNG 页面图。
+
+<粘贴课程大纲>
+```
+
+## 只做规划，不生图
+
+```text
+Use $ian-handdrawn-ppt 先不要生图。
+请把这篇内容规划成一套 10 页左右的中文手绘技术 PPT-style image deck。
+每页给出标题、主旨、版式 archetype、可见文字和图像 brief。
+
+<粘贴素材>
+```
+
+## 技术概念解释图
+
+```text
+Use $ian-handdrawn-ppt 把“RAG 的检索、重排、生成三阶段”做成 3 张 16:9 中文手绘技术解释图。
+每张图保持同一套视觉 shell，避免大标题海报感，Required text only 尽量控制在 8 个短标签以内。
+```
+
+## 根据已有 PPT 提炼重画
+
+```text
+Use $ian-handdrawn-ppt 读取这个 PPTX 的内容和结构，但不要编辑或导出 PPTX。
+请把它重画成一组中文手绘技术 PNG 页面图，保留核心叙事，删掉模板感和长 bullet。
+```

ian-handdrawn-ppt/.gitignore

@@ -0,0 +1,8 @@
+.DS_Store
+__pycache__/
+*.py[cod]
+
+# Local generation outputs
+generated/
+outputs/
+tmp/

ian-handdrawn-ppt/SKILL.md

@@ -0,0 +1,103 @@
+---
+name: ian-handdrawn-ppt
+description: Create final Chinese handdrawn technical article/PPT-style page images from articles, Markdown, PDFs, DOCX files, existing slide decks, course notes, scripts, outlines, or rough ideas. Use when the user asks to turn content into PPT/PPTX/slides/courseware/课件/演示稿/配图/效果图 in a refined Chinese handdrawn technical explanation style, to plan such pages, to choose page layouts from semantic content, or to generate complete image-model pages with Chinese text baked into the final visual. Default article outputs use 21:9 covers and 16:9 body illustrations.
+---
+
+# Ian Handdrawn PPT
+
+Turn source material into a Chinese handdrawn technical PPT-style image deck. Optimize for commercial delivery: clear narrative, semantic page archetypes, exact short Chinese text, refined handdrawn diagrams, and deck-level visual consistency.
+
+## Operating Rule
+
+Default production output is complete raster page images generated by the built-in image generation model. Blog/article covers default to 21:9; body illustrations and standard deck pages default to 16:9. Each page is a final visual deliverable with both diagram and Chinese text included in the image.
+
+In this skill, "PPT", "slides", "page", and "deck" mean finished PPT-style visual page images, not editable presentation/document files. Do not route to presentation or document packaging merely because the user says PPT/PPTX/PDF. Editable PPTX, image-based PPTX, and PDF export are out of scope for this skill.
+
+Do not use deterministic drawing scripts, HTML, SVG, canvas, or python-pptx as the primary visual generator for style samples. Deterministic post-processing is allowed for crop/resize, contact sheets, or exact text overlay when the generated image direction is accepted but image text fidelity is not good enough.
+
+Produce a planning blueprint when the user asks to plan, outline, or design first. Production output is final PNG page images plus a contact sheet when there are multiple pages.
+
+## Resource Map
+
+Load only the references needed for the current task:
+
+- `references/intake.md`: input types, missing-information checks, and concise clarification rules.
+- `references/narrative-planning.md`: deck type selection and story structures.
+- `references/slide-archetypes.md`: semantic mapping from content type to slide layout.
+- `references/visual-dna-v6.md`: handdrawn Chinese technical PPT visual system and deck-level style lock.
+- `references/output-quality.md`: output contracts and verification gates for final image pages.
+- `references/prompt-patterns.md`: prompt templates for complete image-model slide pages.
+
+Use `assets/theme-tokens.json` as the compact theme token file when writing prompts.
+Use `assets/reference-handdrawn-article-illustration-style.png` as the active style anchor for blog/article cover and body illustrations.
+When the current image tool supports local reference images, load or attach this style anchor before generation. When it does not, use the theme tokens plus the reference-match clause in `references/prompt-patterns.md`, and report the style as prompt-matched rather than image-referenced.
+Do not use legacy bordered PPT reference images unless the user explicitly asks to recreate the older bordered look.
+
+## Workflow
+
+1. **Ingest material**
+   - Read the provided content or attached file.
+   - If the input is `.pptx`, extract the source content and visual intent only; do not edit or package the PPTX inside this skill.
+   - If the input is `.docx`, use the Documents plugin or skill to extract content.
+   - If the input is `.pdf`, use the PDF/document tooling appropriate to the task.
+   - If the input is plain text, Markdown, notes, or an outline, parse it directly.
+
+2. **Run intake and gap diagnosis**
+   - Read `references/intake.md` when the content is broad, incomplete, or the output requirements are unclear.
+   - Determine topic, audience, scenario, target length, and source sufficiency.
+   - Ask at most 1-3 questions only when the missing information materially changes the deck.
+   - If no critical information is missing, proceed with reasonable defaults.
+
+3. **Plan the deck narrative**
+   - Read `references/narrative-planning.md`.
+   - Classify the deck as teaching, persuasive, report, product explanation, or knowledge-card style.
+   - Create a slide-by-slide spine: each slide must have one main point.
+
+4. **Map each slide to an archetype**
+   - Read `references/slide-archetypes.md`.
+   - Choose slide layouts from the content semantics, not from a fixed template order.
+   - Vary archetypes so the deck has rhythm.
+
+5. **Apply visual DNA**
+   - Read `references/visual-dna-v6.md`.
+   - Use small exact Chinese text, fine handdrawn lines, light pastel marks, few characters, low visual heaviness, large negative space, and one shared master layout language across the whole deck.
+   - Lock cross-page constants before generating: page role, canvas ratio, near-white paper tone, no-border default, page number, title treatment, title optical size, line weight, pastel palette, corner grid marks, character policy, and spacing rhythm.
+   - Keep the outer shell fixed across pages: page number location, title block, underline, paper tone, corner marks, and visual scale. Vary only the middle semantic diagram area according to the content.
+   - Treat blog/article visuals as two role-specific outputs by default: a 21:9 cover image and 16:9 body illustrations. Do not let body pages look like cover pages.
+
+6. **Build output**
+   - For planning-only requests, deliver a structured blueprint with deck type, slide count, title, main point, archetype, content blocks, visual brief, and missing inputs.
+   - For production requests, use the built-in image generation model to generate one complete page image per slide/visual. Use one image generation call per distinct page brief, not a generic repeated template.
+   - When the user asks for a cover plus body illustrations, generate the cover as 21:9 and body illustrations as 16:9 unless the user specifies otherwise.
+   - Before generating multiple pages, write one compact deck style lock and reuse it verbatim in every page prompt. Add page-specific layout instructions only for the central diagram/content area.
+   - Keep all visible Chinese text short and exact in the prompt. Include a `Required text only` list for each page.
+   - If exact Chinese text is mission-critical or repeated generations render text incorrectly, reduce the text budget first. If needed, generate the accepted visual with blank label spaces and add exact text as deterministic post-processing; the final deliverable is still a raster page image.
+   - Save final selected images into the workspace, and make a contact sheet when generating multiple pages.
+   - Check actual image dimensions. If the image model returns near-target native sizes, report the actual size; normalize body illustrations to 1920x1080 and cover images to 2520x1080 only when strict delivery dimensions are requested.
+
+7. **Verify**
+   - Read `references/output-quality.md`.
+   - Check content accuracy, slide rhythm, Chinese text accuracy, visual consistency, style-anchor match, and commercial handoff readiness.
+   - If verification fails, revise before final delivery.
+
+## Defaults
+
+Use these defaults unless the user says otherwise:
+
+- Language: Simplified Chinese.
+- Audience: Chinese learners with some technical curiosity but not necessarily expert depth.
+- Deck length: 8-12 slides for an article, 15-30 slides for a course module, 5-8 slides for a short idea.
+- Output: final PNG page images plus a contact sheet and short slide blueprint summary.
+- Blog/article visual split: cover image is 21:9; body illustrations are 16:9.
+- Style: refined near-white Chinese handdrawn technical article/PPT illustration V6.
+
+## Final Response
+
+When finished, report:
+
+- The created image folder and contact sheet path.
+- The page count and deck type.
+- Any important assumptions.
+- Verification performed and any remaining risks.
+
+For planning-only outputs, provide the blueprint directly and identify the critical questions to answer before deck production.

ian-handdrawn-ppt/assets/theme-tokens.json

@@ -0,0 +1,90 @@
+{
+  "name": "ian-handdrawn-ppt-v3-article-first",
+  "language": "zh-CN",
+  "canvas": {
+    "cover_ratio": "21:9",
+    "cover_recommended_px": "2520x1080",
+    "body_ratio": "16:9",
+    "body_recommended_px": "1920x1080",
+    "presentation_ratio": "16:9",
+    "output": "complete raster page image",
+    "recommended_px": "1920x1080",
+    "diagram_reference_ratio": "small semantic diagram inside wide margins",
+    "body_diagram_occupancy": "50-60% page width and 35-45% page height",
+    "cover_diagram_occupancy": "50-55% page width"
+  },
+  "colors": {
+    "paper": "#FBFAF5",
+    "ink": "#111111",
+    "muted_ink": "#5F5A50",
+    "blue": "#D9E8F6",
+    "green": "#DCEAD6",
+    "peach": "#F5DEB8",
+    "purple": "#E4DCF4",
+    "accent": "#D66F4D"
+  },
+  "master_layout": {
+    "border": "none by default; only use a very faint inset border when explicitly requested",
+    "page_number": "small upper-left, e.g. 04 / 10",
+    "title": "centered medium restrained handwritten Chinese with pale blue underline",
+    "title_scale_rule": "body-page titles must keep the same optical size even when short",
+    "subtitle": "small centered subtitle beneath title",
+    "corner_marks": "very faint pale grey grid, dots, ruler ticks, or construction marks only",
+    "spacing": "large margins and calm negative space",
+    "variation_policy": "fixed outer shell across pages; vary only central semantic diagram area"
+  },
+  "text_budget": {
+    "title": "5-12 Chinese characters preferred",
+    "subtitle": "3-12 Chinese characters or three short terms separated by ·",
+    "main_labels": "2-5 labels per slide",
+    "captions": "0-6 short captions, usually 2-6 Chinese characters each"
+  },
+  "style": {
+    "line": "fine handdrawn near-black stroke",
+    "objects": "detailed object-based diagrams with pencil hatching",
+    "fills": "very light marker-like pastel fills behind labels or tiny emphasis",
+    "characters": "prefer 0; at most 1 tiny reader-proxy character far in a corner",
+    "text": "short exact Chinese text baked into the final image",
+    "props": "blank or simple line marks unless text is explicitly required",
+    "avoid": [
+      "full-page border",
+      "yellow paper",
+      "beige paper",
+      "large central illustrations",
+      "large colorful cards",
+      "oversized body-page titles",
+      "heavy bottom quote boxes",
+      "shadows",
+      "gradients",
+      "neon colors",
+      "brush calligraphy",
+      "cute poster style",
+      "giant fonts",
+      "long paragraphs",
+      "many people",
+      "script-generated style samples",
+      "incorrect or filler Chinese text"
+    ]
+  },
+  "delivery": {
+    "default": "PNG page images plus contact sheet",
+    "article_cover": "21:9 PNG, recommended 2520x1080",
+    "article_body": "16:9 PNG, recommended 1920x1080",
+    "out_of_scope": "editable PPTX, image-based PPTX, PDF export, and presentation/document packaging",
+    "dimension_policy": "report native image dimensions honestly; normalize body illustrations to 1920x1080 and cover images to 2520x1080 only when strict delivery size is requested",
+    "text_fidelity_policy": "try shorter required text and regeneration first; if exact Chinese remains wrong, use deterministic text overlay on the accepted generated base image"
+  },
+  "style_anchor": "assets/reference-handdrawn-article-illustration-style.png",
+  "slide_archetypes": [
+    "cover-metaphor",
+    "single-concept",
+    "left-right-contrast",
+    "horizontal-process",
+    "circular-mechanism",
+    "branching-map",
+    "classification-map",
+    "matrix-table",
+    "main-metaphor",
+    "takeaway"
+  ]
+}

ian-handdrawn-ppt/references/intake.md

@@ -0,0 +1,78 @@
+# Intake And Gap Diagnosis
+
+Use this reference when source material or output requirements are incomplete.
+
+## Accepted Inputs
+
+- Article, essay, newsletter, blog post
+- Markdown notes
+- DOCX document
+- PDF document
+- Existing PPTX deck
+- Course outline
+- Lesson transcript or video script
+- Rough ideas or bullet points
+- A theme plus desired audience
+
+## Extract First
+
+From the source, identify:
+
+- Topic: what the deck is about
+- Audience: who needs to understand it
+- Scenario: teaching, sales, report, workshop, social content, internal training
+- Goal: what the audience should believe, understand, or do after viewing
+- Core claims: the few statements the deck must preserve
+- Concepts: terms that need explanation
+- Flow: steps, phases, mechanisms, loops, or sequences
+- Contrasts: old/new, wrong/right, before/after, option A/B
+- Evidence: examples, numbers, cases, quotes, observations
+- Risks: claims that need user confirmation
+
+## Defaults
+
+Use these defaults when safe:
+
+- Audience: Chinese learners with moderate technical familiarity
+- Scenario: teaching or public explanation
+- Length: 8-12 slides for one article; 15-30 for a course module
+- Depth: practical conceptual explanation, not academic proof
+- Style: V6 handdrawn Chinese technical PPT
+- Output: role-specific final PNG page images when the user asks for slides/effect images; blueprint when the user asks for planning
+- Blog/article visual split: 21:9 cover image plus 16:9 body illustrations when the user asks for both cover and body images
+
+Interpret "PPT" as PPT-style final page images by default. Editable PPTX, image-based PPTX, PDF export, and object-level presentation modification are out of scope for this skill.
+
+## Ask Only For Critical Missing Inputs
+
+Ask at most 1-3 concise questions when the answer changes the deck materially.
+
+Ask when:
+
+- The topic is ambiguous.
+- Audience depth would change the language level.
+- Use scenario is unknown and affects structure.
+- The user asks for commercial delivery but gives only a rough topic.
+- Source material contains claims that look important but unsupported.
+- The user gives a strict page count or delivery format but it is missing.
+
+Do not ask when:
+
+- You can choose a reasonable default and continue.
+- The missing detail only affects polish, not structure.
+- The user clearly wants a first draft.
+
+## Gap Diagnosis Output
+
+When you need to ask, frame the default and the missing choice:
+
+```text
+我可以先按“中文技术课件，面向有基础的新手，10 页左右”处理。
+但有一个决定性信息会影响结构：这套 PPT 是用于教学、销售说明，还是公开分享？
+```
+
+## Source Sufficiency Levels
+
+- **Enough**: has thesis, several points, examples or process, and a conclusion.
+- **Thin but usable**: has thesis and points but few examples. Proceed and mark assumptions.
+- **Insufficient**: only a theme or slogan. Ask for goal, audience, and 3-5 raw points.

ian-handdrawn-ppt/references/narrative-planning.md

@@ -0,0 +1,98 @@
+# Narrative Planning
+
+Choose the deck story before designing slides. The same content can produce different decks depending on the scenario.
+
+## Deck Types
+
+### Teaching Deck
+
+Use for courses, tutorials, explainers, workshops.
+
+Default spine:
+
+1. Why this matters
+2. The common confusion
+3. Core concept
+4. Mechanism or structure
+5. Worked example
+6. Mistakes or edge cases
+7. Practical method
+8. Summary
+
+### Persuasive Deck
+
+Use for opinion essays, method pitches, strategy arguments.
+
+Default spine:
+
+1. Existing pain
+2. Why old framing fails
+3. New judgment
+4. Evidence or examples
+5. Framework
+6. What changes in practice
+7. Risks or objections
+8. Call to action or conclusion
+
+### Report Deck
+
+Use for project updates, internal reviews, proposals.
+
+Default spine:
+
+1. Context
+2. Goal
+3. Current state
+4. Key finding
+5. Options or plan
+6. Risks
+7. Next steps
+
+### Product Explanation Deck
+
+Use for tools, systems, features, or technical products.
+
+Default spine:
+
+1. User pain
+2. System idea
+3. How it works
+4. Main workflow
+5. Differentiators
+6. Example scenario
+7. Adoption or next step
+
+### Knowledge Card Deck
+
+Use for short social content, mini lessons, or lightweight summaries.
+
+Default spine:
+
+1. One sharp question
+2. Key explanation
+3. Contrast or mechanism
+4. Example
+5. Takeaway
+
+## Slide Planning Rules
+
+- Give each slide one main point.
+- Prefer a clear narrative over exhaustive coverage.
+- Convert long paragraphs into structure, not bullet piles.
+- Use examples when the source includes them; do not invent factual case studies.
+- If an example is needed but absent, either ask or label it as a hypothetical example.
+- Vary slide archetypes to avoid repeated layouts.
+
+## Slide Blueprint Fields
+
+For each slide, produce:
+
+- Slide number
+- Title
+- Main point
+- Archetype
+- Content blocks
+- Visual brief
+- Speaker note, if useful
+- Source support or assumption
+

ian-handdrawn-ppt/references/output-quality.md

@@ -0,0 +1,127 @@
+# Output And Quality Gates
+
+Use this before final delivery.
+
+## Output Modes
+
+### Blueprint
+
+Use when the user wants planning first or the source is not ready for production.
+
+Include:
+
+- Deck type
+- Assumptions
+- Slide count
+- Slide-by-slide blueprint
+- Missing inputs or risks
+
+### Final Image Deck
+
+Use by default when the user asks for PPT/PPTX/slides/courseware/配图/效果图 as finished visual page images.
+
+PPT/PPTX wording alone does not mean editable output. Treat it as PPT-style final PNG page images. Editable PPTX, image-based PPTX, PDF export, and object-level presentation modification are out of scope for this skill.
+
+Deliver:
+
+- Role-specific PNG pages:
+  - Blog/article cover: 21:9, preferably 2520x1080
+  - Body illustrations and standard pages: 16:9, preferably 1920x1080
+- Contact sheet for quick review when there are multiple pages
+- Short blueprint summary
+- Assumptions and verification
+
+### Polished Delivery
+
+Use when the user asks for polished delivery.
+
+Deliver when feasible:
+
+- Final PNG pages
+- Contact sheet
+- Speaker notes
+- Visual style notes
+
+## Content Quality Gate
+
+Check:
+
+- Each slide has one main point.
+- Slide order creates a coherent story.
+- Claims are supported by source material or clearly marked as assumptions.
+- Terminology is consistent.
+- Examples are not fabricated as facts.
+- The audience level is appropriate.
+
+## Layout Quality Gate
+
+Check:
+
+- Archetype matches slide semantics.
+- Layouts vary by content semantics, but share one master visual language.
+- Non-cover titles are not oversized.
+- Short body-page titles are not enlarged into cover headlines.
+- Central body diagrams stay compact, usually 50-60% page width and 35-45% page height.
+- Slides are not just bullet lists.
+- Diagrams explain, not decorate.
+
+## Visual Quality Gate
+
+Check:
+
+- Small readable Chinese text
+- Fine handdrawn lines
+- Light pastel marks
+- Few characters
+- Near-white warm paper, not yellow or beige.
+- No full-page border unless explicitly requested.
+- Faint corner construction marks only; no heavy frame.
+- Page number, centered title, and understated underline are consistent when the deck uses the reference style.
+- Pages feel like one illustrator and one deck, not unrelated images.
+- No large card-heavy look
+- No poster-like full-page illustration look
+- No heavy bottom quote/conclusion boxes
+- No shadows, gradients, neon, or corporate template feel
+
+## Dimension Gate
+
+Check:
+
+- Page images match their role:
+  - Cover image: 21:9 or very close.
+  - Body illustration: 16:9 or very close.
+- Actual pixel dimensions are reported honestly.
+- If strict delivery size is requested, normalize accepted body images to 1920x1080 and cover images to 2520x1080 after generation with high-quality resampling, then verify the normalized images still look sharp.
+- Do not claim 1920x1080 or 2520x1080 if the generated source files are a different native size.
+
+## Image Text Gate
+
+Check:
+
+- Required Chinese text appears and is readable.
+- Text is short enough for the image model to render cleanly.
+- No fake filler text, random English, URLs, watermarks, or extra labels.
+- Props such as books, screens, documents, and cards are blank or only contain line marks unless their text is explicitly listed.
+- If exact text fails, reduce text and regenerate.
+- If exact text remains wrong after simplification, use deterministic text overlay on the accepted generated base image, then re-check readability, alignment, and style consistency before delivery.
+
+## Contact Sheet Gate
+
+For multiple images, make a contact sheet and inspect it before delivery:
+
+- Backgrounds match and are not drifting yellow.
+- Page shells match: no-border default, same title position, same underline style.
+- Body-page titles look optically equal even when character counts differ.
+- Diagrams have similar visual weight and do not suddenly become huge.
+- The cover is clearly 21:9 and body illustrations are clearly 16:9.
+
+## Final Report
+
+Mention:
+
+- Output folder and contact sheet path
+- Page count
+- Deck type
+- Major assumptions
+- Verification performed
+- Remaining risks

ian-handdrawn-ppt/references/prompt-patterns.md

@@ -0,0 +1,197 @@
+# Prompt Patterns
+
+Use these patterns when generating final image-model PPT pages or planning slide content.
+
+## Deck Style Lock
+
+Write this once before generating a multi-page deck. Reuse the same style language in every page prompt.
+
+```text
+Refined commercial Chinese handdrawn technical article/PPT illustration.
+Complete raster image on very light warm white paper, near #FBFAF5, with extremely subtle grain.
+No full-page border and no rectangular frame unless explicitly requested.
+Upper-left small page number in handwritten style.
+Centered medium Chinese title with one pale blue handdrawn underline.
+Small subtitle under title when needed.
+For body pages, keep title size optically consistent across pages; do not enlarge short titles.
+Fine black ink and pencil linework, delicate hatching, stable but slightly irregular.
+Muted pastel marker labels: pale blue, sage green, peach, lavender.
+Sparse corner construction marks only: faint pale grey grid, dots, ruler ticks, measurement lines.
+Generous negative space, calm premium teaching-note feeling.
+Mostly small object-based diagrams; no more than one tiny person, placed far in a corner.
+Props should be blank or contain only simple line marks unless their text appears in Required text only.
+Avoid full-page border, yellow paper, beige paper, giant fonts, cheap poster look, childish doodles, many characters, thick marker strokes, dense bullets, corporate template style, shadows, gradients, neon, watermark, gibberish text, English filler.
+```
+
+## Reference Match Clause
+
+Use this when the user wants the reference-image look or a commercial style sample.
+
+```text
+Match the approved article-illustration shell: near-white warm paper, no full-page border, upper-left small page number, centered restrained handwritten Chinese title, one pale blue underline, small subtitle beneath, sparse corner grid/dot construction marks, fine ink-and-pencil object drawings with delicate hatching, small refined central diagram, and large negative space. Keep the outer shell consistent across pages; only vary the central diagram layout according to the content.
+```
+
+## Page Role Locks
+
+Use exactly one page role in every production prompt.
+
+```text
+Page role: cover image.
+Canvas: 21:9 ultra-wide, preferred final size 2520x1080.
+Title may be medium-large but still elegant and restrained.
+Main metaphor occupies about 50-55% of page width.
+No page number unless requested.
+```
+
+```text
+Page role: body illustration.
+Canvas: 16:9, preferred final size 1920x1080.
+This is not a cover page.
+Title is medium and restrained, optically consistent with the other body pages.
+Even if the title is short, do not enlarge it.
+Central diagram occupies about 50-60% of page width and 35-45% of page height.
+```
+
+## Complete Page Image Prompt
+
+Use for each final page image. Use the built-in image generation model directly. Do not substitute script-generated layouts for style trials.
+
+```text
+Use case: productivity-visual.
+Asset type: one complete Chinese handdrawn technical article/PPT image, final raster page.
+Preferred final size: use the page role target, 2520x1080 for 21:9 cover or 1920x1080 for 16:9 body illustration when supported; otherwise keep native role ratio and report actual dimensions.
+
+Create page <number>/<total> of a coherent deck.
+Page role: <cover image | body illustration | standard deck page>
+Page number text exactly: <NN / TT, omit for cover if page number is not wanted>
+Title exactly: <short Chinese title>
+Subtitle exactly: <optional short Chinese subtitle>
+Archetype: <cover metaphor | left-right contrast | horizontal process | circular mechanism | branching map | classification map | matrix table | main metaphor | takeaway>
+Main point: <one sentence>
+
+Apply the deck style lock: <paste compact style lock>.
+Reference match clause: <paste when needed>.
+
+Composition:
+<specific layout based on semantics. Describe object-based handdrawn diagram, not generic boxes.>
+Scale lock:
+<for body illustration: central diagram 50-60% page width and 35-45% page height; title same optical size as other body pages.>
+
+Required text only:
+<list every visible Chinese text item exactly. Keep this list short.>
+
+Avoid:
+full-page border, yellow paper, beige paper, oversized central objects, oversized body-page title, heavy bottom boxes, extra text, invented micro-labels, gibberish, English, watermark, crowded composition, many people, childish cartoons, thick outlines, saturated colors, corporate template look.
+```
+
+## 21:9 Cover Image Prompt
+
+Use when the user asks for a blog/article cover.
+
+```text
+Use case: Chinese blog/article cover image.
+Asset type: one complete 21:9 ultra-wide Chinese handdrawn technical cover image, final raster page.
+Preferred final size: 2520x1080 if supported.
+
+Page role: cover image.
+Title exactly: <title>
+Subtitle exactly: <subtitle>
+Archetype: cover metaphor
+Main point: <one sentence>
+
+Apply the deck style lock: <paste compact style lock>.
+Apply the cover role lock: <paste cover role lock>.
+
+Composition:
+<one small refined central metaphor, occupying about 50-55% page width, with wide empty margins.>
+
+Required text only:
+<short exact visible Chinese text list>
+
+Avoid:
+full-page border, yellow paper, beige paper, large poster composition, giant title, heavy boxes, extra text, invented micro-labels, English, gibberish, watermark.
+```
+
+## 16:9 Body Illustration Prompt
+
+Use for article body images and regular explanatory pages.
+
+```text
+Use case: Chinese blog/article body illustration.
+Asset type: one complete 16:9 Chinese handdrawn technical body illustration, final raster page.
+Preferred final size: 1920x1080 if supported.
+
+Page role: body illustration.
+Create page <number>/<total> of a coherent illustration set.
+Page number text exactly: <NN / TT>
+Title exactly: <title>
+Subtitle exactly: <subtitle>
+Archetype: <left-right contrast | horizontal process | circular mechanism | branching map | classification map | matrix table | main metaphor | takeaway>
+Main point: <one sentence>
+
+Apply the deck style lock: <paste compact style lock>.
+Apply the body illustration role lock: <paste body role lock>.
+
+Composition:
+<small refined central semantic diagram. Keep the diagram about 50-60% page width and 35-45% page height.>
+
+Required text only:
+<short exact visible Chinese text list>
+
+Avoid:
+full-page border, yellow paper, beige paper, large central objects, oversized title, heavy bottom quote box, extra text, invented micro-labels, English, gibberish, watermark, crowded composition.
+```
+
+## Compact Three-Station Body Page
+
+Use when the content is a process, pipeline, or three-part framework in the approved article-illustration style.
+
+```text
+Composition: three calm stations across the middle with slim arrows between them.
+Each station has a pastel marker label above, a detailed handdrawn object in the center, and one tiny semantic caption below with a short colored underline.
+Keep object drawings more detailed than icons: pencil hatching, fine contour lines, small construction details.
+Leave wide margins and quiet negative space.
+Keep the three-station group compact: about 55-60% page width on body illustrations.
+No full-page border. Near-white warm paper. Faint corner marks only.
+Use at most one tiny figure in the lower-right corner.
+```
+
+## Slide Content Compression
+
+Use this prompt internally:
+
+```text
+Compress this source section into one slide.
+Keep exactly one main point.
+Choose the best semantic archetype.
+Return a title, 2-4 content blocks, optional annotations, required visible text, and a complete page image brief.
+Do not produce a bullet dump.
+```
+
+## Multi-Page Consistency Pass
+
+Before generating page images, check:
+
+```text
+- Does every page share the same page number position?
+- Does every page use the same near-white paper, no-border shell, title underline, line weight, and pastel family?
+- Are cover images 21:9 and body illustrations 16:9?
+- Are body-page titles optically the same size, including short titles?
+- Is the central diagram compact enough, or did the model fill the page?
+- Are layouts semantic rather than randomly different?
+- Are there too many people across the deck?
+- Is each page's Required text only list short enough for clean image rendering?
+- Are props free of fake English, URLs, and filler text?
+```
+
+## Text Fidelity Fallback
+
+Use only after the image direction is accepted and image-generated Chinese text remains incorrect.
+
+```text
+Regenerate or reuse the accepted visual with blank reserved spaces for every exact Chinese label.
+Do not invent any placeholder writing.
+Leave clean pale marker labels or empty paper areas where text will be overlaid.
+After generation, add only the Required text items with deterministic local post-processing, matching the small hard-pen Chinese title/body style as closely as practical.
+Re-check that no duplicate, fake, or misspelled generated text remains visible.
+```

ian-handdrawn-ppt/references/slide-archetypes.md

@@ -0,0 +1,142 @@
+# Slide Archetypes
+
+Select layout by semantic content. Do not use a fixed template order.
+
+## Archetype Map
+
+| Content Semantic | Archetype | Use When |
+|---|---|---|
+| Opening theme | Cover metaphor | A deck or chapter starts with one strong concept |
+| Definition | Single concept explainer | One term or idea needs unpacking |
+| Misunderstanding vs correct framing | Left-right contrast | The source contrasts two ways of thinking |
+| Before vs after | Transformation contrast | The content shows a shift or upgrade |
+| Steps or pipeline | Horizontal process | There is a sequence from input to output |
+| Feedback loop | Circular mechanism | The process repeats or learns over time |
+| Decision conditions | Branching map | The audience must choose among cases |
+| Taxonomy | Classification map | The content groups concepts or types |
+| Multi-axis comparison | Matrix table | There are options compared across dimensions |
+| Abstract system | Main metaphor diagram | A metaphor explains the structure better than boxes |
+| Warning or caveat | Annotation slide | A risk needs focused attention |
+| Summary | Takeaway slide | The deck needs a final remembered judgment |
+
+## Core Archetypes
+
+### Cover Metaphor
+
+Use for a deck/chapter opening.
+
+Structure:
+
+- Medium-to-large Chinese title
+- One main handdrawn metaphor
+- Optional subtitle
+- Optional small reader character
+
+Only cover slides may use relatively large type.
+
+### Single Concept Explainer
+
+Use for definitions and conceptual pivots.
+
+Structure:
+
+- Title
+- Central concept label
+- Two to four supporting annotations
+- Small icon or metaphor
+
+### Left-Right Contrast
+
+Use for wrong/right, old/new, dirty/clean, before/after.
+
+Structure:
+
+- Two main groups
+- Light divider or arrow
+- Left can look messier; right can look more structured
+- One small conclusion
+
+### Horizontal Process
+
+Use for stages, pipelines, execution flows.
+
+Structure:
+
+- Thin line from left to right
+- Four to seven small modules
+- Light arrows
+- One or two annotation boxes
+
+### Circular Mechanism
+
+Use for loops, agents, feedback, iteration.
+
+Structure:
+
+- Three to six loop stages
+- Curved arrows
+- Center label for the mechanism
+- Edge notes for exit conditions or risks
+
+### Branching Map
+
+Use when a decision depends on task type or condition.
+
+Structure:
+
+- Central question or decision
+- Three to five branches
+- Each branch has a short condition and action
+
+### Classification Map
+
+Use for frameworks and conceptual families.
+
+Structure:
+
+- Parent concept
+- Three to five branches
+- One to three micro labels under each branch
+
+### Matrix Table
+
+Use for comparisons.
+
+Structure:
+
+- Handdrawn grid
+- Short row/column labels
+- Small icons in selected cells
+- Optional reader character at edge
+
+### Main Metaphor Diagram
+
+Use for hard abstract ideas.
+
+Structure:
+
+- One large metaphor object or scene
+- Thin arrows and labels around it
+- Minimal small characters
+
+### Takeaway Slide
+
+Use for final summary.
+
+Structure:
+
+- One clear judgment
+- Three or fewer supporting echoes
+- Small metaphor or icon
+
+Avoid turning this into a motivational poster.
+
+## Anti-Patterns
+
+- Reusing one layout for every slide
+- Turning every idea into a huge card
+- Putting a character next to every node
+- Oversized title on non-cover slides
+- Full-slide bullet lists
+- Tiny decorative elements without semantic function
+

ian-handdrawn-ppt/references/visual-dna-v6.md

@@ -0,0 +1,167 @@
+# Visual DNA V6
+
+Use this visual system for Chinese handdrawn technical PPT image pages.
+
+## Positioning
+
+Create refined handdrawn technical explanation images: small exact Chinese text, fine ink lines, light pastel marks, sparse characters, clear structure, and a premium Chinese article/teaching-note feeling.
+
+Do not create generic business templates, cute slide decks, big-title quote cards, dense whiteboards, or poster-like full-page illustrations.
+
+## Canvas
+
+- Final output is a complete raster image.
+- Use role-specific canvases:
+  - Blog/article cover image: 21:9, preferably 2520x1080.
+  - Body illustration: 16:9, preferably 1920x1080.
+  - Standard deck page: 16:9 unless the user specifies otherwise.
+- Use very light warm white paper, near `#FBFAF5`, with only extremely subtle grain. Avoid yellow, beige, old paper, or tea-stained paper.
+- Default to no full-page border. Do not draw a rectangular frame around the page unless the user explicitly asks for a card/bordered look.
+- Keep only tiny notebook marks at the corners: faint grey grid, dots, ruler ticks, or construction lines. These marks should feel incidental and very quiet.
+- Keep the central diagram small and refined:
+  - Body illustration diagram: about 50-60% of page width and 35-45% of page height.
+  - 21:9 cover metaphor: about 50-55% of page width.
+- Use large negative space. The page should feel like a premium article illustration, not a filled PPT canvas.
+
+## Master Consistency
+
+Lock these constants across a multi-page deck before generating:
+
+- Same paper tone and grain.
+- Same no-border shell by default; if a border is explicitly requested, use the same very faint border style across all pages.
+- Same small page number convention, usually `01 / 10` in the upper-left.
+- Same title treatment: centered, hard-pen Chinese, medium size, with one short pale blue underline.
+- Same subtitle treatment: one short line below the title, smaller than title.
+- Same title optical size across body pages. Do not enlarge short titles just because they contain fewer characters.
+- Same diagram line weight and hatching style.
+- Same pastel label fills: light blue, sage green, pale peach, soft lavender.
+- Same character policy: usually none; at most one tiny reader/engineer far in a corner.
+- Same margin rhythm: wide margins, calm center, no full-bleed crowding.
+
+If multiple pages feel like different illustrators made them, revise prompts before regenerating.
+
+## Deck Shell And Variation
+
+Treat a deck as one fixed shell plus variable semantic diagrams:
+
+- Fixed shell: paper tone, grain, no-border default, upper-left page number, centered title, pale blue title underline, subtitle position, corner construction marks, wide margins, and overall text scale.
+- Variable middle: the object diagram, flow direction, grouping, arrows, labels, and metaphor chosen for the slide's content.
+- Vary layouts by semantic archetype, not by changing the deck identity. A horizontal process, contrast page, loop, and layered map can feel different while still sharing the same shell.
+- Do not create variety by moving the title, adding/removing borders, switching font mood, adding more people, changing title scale, or changing the palette.
+
+## Typography
+
+The final image contains the visible Chinese text. Keep the text short, exact, and manually checkable.
+
+Style:
+
+- Heading: clear handdrawn hard-pen Chinese title feel
+- Body: small readable hard-pen Chinese
+- Avoid calligraphic brush style, dramatic flyaway strokes, childish doodle fonts, and heavy advertising type.
+
+Scale guidance:
+
+- Cover title: medium-large, but only on cover pages
+- Body illustration title: medium and restrained
+- Normal slide title: medium
+- Section labels: small-medium, often inside small pastel marker labels
+- Node text: small
+- Annotations: smaller
+- Bottom takeaway: only slightly stronger than body text
+
+Avoid huge typography. The page should feel authored and precise, not like a cheap poster.
+Short-title trap: when a title has only 4-6 Chinese characters, explicitly keep the same optical size as other body pages. Do not let the image model make it a cover headline.
+
+## Text Budget
+
+For reliable Chinese rendering, prefer:
+
+- Title: 5-12 Chinese characters.
+- Subtitle: 3-12 Chinese characters, or three short terms separated by `·`.
+- Main labels: 2-5 labels per slide.
+- Captions/annotations: 0-6 short items, each usually 2-6 Chinese characters.
+
+If a slide needs more text than this, split it or move detail into speaker notes/adjacent pages.
+
+## Color
+
+- Background: very light warm white paper, near `#FBFAF5`
+- Line: near black, fine handdrawn stroke
+- Pastels: very light blue, sage green, peach, lavender
+- Accent: small red-orange star, dot, or underline only when useful
+
+Avoid:
+
+- Large saturated blocks
+- Shadows
+- Gradients
+- Neon colors
+- Product-card styling
+
+## Line And Shape
+
+- Use fine handdrawn lines.
+- Lines should be stable but slightly irregular.
+- Use careful line-art objects: nets, sieves, drawers, cards, cabinets, funnels, clocks, dials, trays, shelves, documents, magnifying glasses.
+- Use small rounded boxes, tags, speech bubbles, dashed notes, arrows, tables, and diagrams only when they explain the content.
+- Most containers should be paper-filled with thin outline; use pastel fills sparingly.
+- Prefer detailed object drawing and hatching over flat vector icons.
+- Arrows should be slim and quiet.
+- Props such as cards, books, screens, trays, and documents should be blank or contain only simple line marks unless the text is listed in `Required text only`. Avoid fake English, fake URLs, and filler writing.
+
+## Characters
+
+Characters are reader proxies or metaphor actors.
+
+Rules:
+
+- Prefer no character. If used, place one tiny character in a corner or edge, not beside every node.
+- Max one tiny emotional character per page.
+- Do not place a character beside every node.
+- Character area should stay visually small.
+
+## Slide Density
+
+Prefer:
+
+- Two to four main structure groups
+- Four to twelve micro modules
+- Zero to three annotations
+- One main idea per slide
+
+Avoid:
+
+- Three to five huge colored cards
+- Oversized bottom quotes
+- Heavy bottom conclusion boxes
+- Overcrowded bullet text
+- Decorative people or icons that do not explain anything
+
+## Text In Images
+
+For final image-page decks:
+
+- Chinese text is part of the final generated page.
+- Keep required text short: title, subtitle, section labels, and a few micro labels.
+- Always provide a `Required text only` list in the image prompt.
+- Avoid long paragraphs, dense bullets, URLs, code blocks, and any non-essential text.
+- If text accuracy is poor, regenerate with fewer words or split the page.
+- If exact Chinese text is still wrong after simplification, keep the accepted visual direction and add the required text with deterministic post-processing. This is a text-fidelity repair, not the primary visual-generation method.
+
+## Reference-Style Anchor
+
+For blog/article cover and body illustrations, use `assets/reference-handdrawn-article-illustration-style.png` as the active style anchor:
+
+- Near-white warm paper, not yellow.
+- No full-page border.
+- Faint corner grid/dot construction marks only.
+- Small, refined central object diagrams with large negative space.
+- Medium restrained body-page titles with short pale blue underlines.
+- Fine ink-and-pencil object drawings with delicate hatching.
+- No extra text beyond the required visible Chinese text.
+
+Legacy bordered PPT references are archived outside active assets. Do not load or imitate them unless the user explicitly asks for the older bordered page style.
+
+## Final Look
+
+The page should feel like a technical author carefully drew a compact concept illustration on near-white paper, then polished it enough for paid teaching, article publishing, or commercial delivery.