Xiaobei Skill Academic Paper To Ppt

小北在读研 · Academic Paper to PPT

Core Behavior

When this skill is triggered, execute the workflow in this exact order:

1. Analyze the uploaded source document
2. If the user provides a PPT/PPTX template, analyze the template's visual style and page structure
3. Create ppt_outline.md
4. Generate one complete visual preview scheme based strictly on ppt_outline.md using $imagegen, then immediately display that image inline in the chat
5. Generate every slide page as a full-page image using $imagegen; immediately display each completed page image inline in the chat before waiting for the whole deck
6. Create imagegen_manifest.json recording the selected $imagegen generation method and every generated asset
7. Assemble the generated slide images into a .pptx file where each image fills one complete slide
8. Validate that required output files exist and match the imagegen manifest

Never skip stages unless the user explicitly requests only one stage.

Strict image-only contract:

• A final slide image means the whole finished PPT page is baked into one bitmap, including Chinese title, body text, labels, diagrams, tables, and page visual design.
• The final .pptx must contain one full-slide image per slide and no editable visible text boxes.
• Background-only slide images are invalid. Do not generate text-free templates, blank layout backgrounds, placeholder panels, blank areas for later text, or editable-text overlays.
• Prompts for final slide images must not include phrases such as text-free, no readable words, background template, blank areas, editable Chinese text, text to be added later, or equivalent wording.
• If $imagegen cannot generate complete page images with the required text, retry with shorter full-page prompts or switch to the other allowed $imagegen mode. If complete page images still cannot be produced, stop at image generation and report the failed stage. Do not assemble a fallback PPT.

Chat-visible image contract:

• Apply this contract to both built-in image_gen and CLI generation. Changing the image backend must not make generated images invisible to the user.
• Stage 2 produces one visual-scheme image. As soon as it exists, show it in the chat with local-image Markdown using its absolute path, for example ![PPT 视觉方案](/absolute/path/visual_preview.png).
• During Stage 3, show each completed slide image individually as soon as it is available. Label it briefly as 第 X/N 页, then render that page with its absolute local path. A text-only progress message such as “已生成 6 页” does not satisfy this requirement.
• Do not wait until all pages finish before showing them. Do not replace per-page visibility with only a contact sheet, folder path, file count, or final PPTX link.
• In CLI mode, run generation in a monitorable process and poll for newly completed files. After each poll, display every newly completed slide_XX.png individually before continuing to wait. Limited concurrency is allowed, but the chat must expose each completed page rather than hiding the batch until completion.
• In built-in mode, the tool result may render automatically; still add the brief 第 X/N 页 label and preserve/copy the selected image into the workspace for manifesting and PPT assembly.
• After all pages are visible, optionally add a complete contact sheet for convenient overview, then provide the final PPTX.

Required image-generation dependency:

• The visual preview and full-slide generation stages must load and follow the installed $imagegen skill.
• Use built-in image_gen first when the user does not specify a generation mode.
• Use CLI mode when the user explicitly requests it or when a real built-in attempt is unavailable or fails, but only through the runtime, endpoint, and credential mechanism authorized in the current user's $imagegen workflow.
• Do not hardcode a maintainer's workstation path, account name, private endpoint, or credential source. Never display, log, or persist credentials.
• In CLI mode, write generated image files under outputs/imagegen/ and record method imagegen_cli_fallback in imagegen_manifest.json.
• Use a Python environment with the packages in requirements.txt for PowerPoint assembly and validation.
• Do not replace $imagegen with textual visual suggestions, SVG/vector/code-native drawings, HTML/CSS/canvas mockups, manual diagrams, or a generic image-generation placeholder.
• When built-in mode is used, copy the selected generated files from $CODEX_HOME/generated_images/... or the workspace into outputs/imagegen/generated_slides/, and record method built_in_image_gen in imagegen_manifest.json.
• If an explicitly selected CLI attempt fails to generate real files, try built-in image_gen unless the user asked for CLI only. If the default built-in attempt fails, use CLI only when the current environment authorizes and supports that fallback.
• If neither CLI fallback nor built-in $imagegen can generate real image files, stop at that stage and report the incomplete stage. Do not assemble a PPT from non-imagegen substitutes.
• A visual preview generated with $imagegen is not enough. Every final slide image recorded in imagegen_manifest.json must also be generated with $imagegen.
• If CLI fallback fails only for long Chinese prompts, that is not permission to generate text-free backgrounds. First shorten each prompt while preserving the requirement that the final image contains the slide title and key visible text. Then try built-in image_gen if needed. Stop rather than replacing text with editable PowerPoint objects.

Do not expand the full Markdown outline in the chat. Do not output long analysis. Do not invent information that is not in the uploaded document. Do not cite or use external sources unless the user explicitly asks for external supplementation.

Visual Style Priority

Use this priority order for the deck's visual style:

1. Explicit style instructions from the user in the current request.
2. A user-provided PPT/PPTX template, when available.
3. The default academic visual preset in references/visual-style-spec.md.

If a template is provided, treat it as a visual and structural reference only:

• Extract its slide size, color palette, typography, title hierarchy, section-page design, content-page design, chart/table treatment, spacing, and visual motifs.
• Prefer the template's cover, directory, section divider, body, chart/table, summary, and closing-page patterns when generating the visual preview and full-slide images.
• Use the template's topic distribution to decide the outline's page grouping and content density, but do not copy its placeholder text as source content.
• Do not force the generated PPT to use every page from the template. Compress or expand the template's page families into a source-appropriate academic or project-report narrative.
• Do not add market analysis, team information, intellectual property, or implementation-condition pages unless those topics are present in the uploaded source document or explicitly requested by the user.
• If the template conflicts with source-document fidelity, source-document fidelity wins.

If no template is provided, use the default academic preset from references/visual-style-spec.md.

Stage 1: Document Parsing and Markdown PPT Outline

Input:

• A user-uploaded literature paper, academic article, research report, technical document, PDF, DOCX, Markdown file, or plain text document.
• Optional: a user-provided PPT/PPTX template used only for visual style, page-family structure, and outline density guidance.

Output:

• Create or update a file named exactly:

ppt_outline.md

The file must contain a complete Markdown PPT outline of about 12-18 pages by default. Use about 10 pages only when the source document is short or the user explicitly asks for a compact deck.

Strict rules:

1. All content must be strictly based on the uploaded document.
2. Do not fabricate information.
3. If information is not explicitly stated in the original document, write: 原文未明确说明.
4. Do not use external sources.
5. Keep the structure suitable for academic presentation, course presentation, thesis sharing, or project introduction.
6. Each slide must include:
   • 页面标题
   • 页面类型
   • 核心内容
   • 内容展开
   • 建议展示方式
   • 视觉生成要点
   • 原文依据
7. Each slide must contain enough detail to support later PPT generation, especially enough visual and content cues for $imagegen.
8. Prioritize extracting important data, experiments, methods, charts, formulas, models, algorithms, case studies, results, limitations, and conclusion evidence from the original document.
9. If the uploaded document is long, reorganize the content into a presentation logic instead of copying the original table of contents.
10. If a PPT template is provided, use its section rhythm and page families to guide page distribution and content density, while keeping all factual content grounded in the source document.
11. The output must be usable by AI PPT tools, Codex, or visual design tools.

Use the structure in references/outline-template.md.

After Stage 1, reply briefly only:

• 已生成 ppt_outline.md
• 文件中包含约 12-18 页 PPT 大纲
• 内容严格基于上传文档

Do not paste the full outline into chat.