Group Meeting Ppt

做组会PPT

Goal

Turn an experiment folder into a presentation-ready package:

• Read the experiment directory and recent result exports deeply.
• Extract the experiment story, methods, data usage, metrics, risks, and conclusions from actual files/code.
• Produce an image plan and AI image prompts in Markdown.
• Generate or collect numbered slide images.
• Build a PPTX with one full-slide image per page.
• Add detailed Chinese speaker notes without corrupting Chinese text.

Workflow

1. Build Source Context

Start from the user-specified directory or current working directory.

Inspect:

• Top-level reports, summaries, README/navigation files, and dated result exports.
• Scripts that define data loading, training, inference, evaluation, and fusion logic.
• Tables, CSV/JSON/NPY exports, logs, and generated figures.
• Existing prompt files, notes files, image folders, and PPT outputs if continuing prior work.

Prefer rg --files, rg, Get-ChildItem, and targeted Get-Content. Do not assume a directory name is literal if the actual folder differs; state the resolved path.

When answering methodology questions, cite actual code behavior:

• Training/inference split and file globs.
• Whether H/P groups are used for unsupervised training, supervised losses, validation, or evaluation.
• Which losses are active, newly introduced, or carried over.
• Which result file supports each reported metric.

2. Extract Presentation Story

Before making visuals, form a concise talk narrative:

• Research problem and why the current experiment exists.
• Data representation and pipeline.
• Baselines and incremental method changes.
• Key quantitative results and negative results.
• What the experiment proves, what it does not prove, and the next technical direction.

For medical/research experiments, distinguish:

• Empirical result vs. interpretation.
• Mainline method vs. failed/negative branch.
• Code-confirmed fact vs. inference from filenames/reports.

3. Create Image Plan And Prompts

Create a Markdown prompt file in the experiment or output folder, usually:

<experiment>_PPT_AI_IMAGE_PROMPTS.md or PROMPTS.md

Use numbered images:

01, 02, ..., with each entry containing:

• Slide purpose.
• What the image should show.
• AI image prompt.
• Optional style constraints.

Do not avoid text/numbers inside prompts if the user explicitly says the current image model can handle them. Still keep important text short and visually large.

Recommended prompt style:

• Use 16:9 wide slide illustration.
• Request clean research-presentation composition.
• Include meaningful labels, arrows, small charts, and schematic anatomy/pipeline elements.
• Avoid relying on exact tiny table values unless needed.
• Keep one core message per image.

4. Generate Or Collect Images

When the user asks to generate images, use the available image generation tool for each prompt. Save/copy the final image assets into a dedicated folder such as:

<experiment>_ppt_ai_images/

Use stable names:

quest003_ppt_image_01.png quest003_ppt_image_02.png ...

Also copy the prompt Markdown into the same folder as PROMPTS.md.

After generation, verify:

• Expected image count exists.
• Numbering is continuous.
• Files are readable image assets.

5. Build PPTX

If asked for a simple PPT, create one slide per image and make the image cover the whole slide.

Use python-pptx for slides-only PPTX when notes are not needed.

For Chinese speaker notes, prefer PowerPoint COM automation on Windows because direct OpenXML or python-pptx notes manipulation can create repair warnings or corrupt Chinese notes.

Recommended output names:

• QuestXXX_PPT_pythonpptx_slides_only.pptx
• QuestXXX_PPT_Final_with_Chinese_Notes.pptx
• Optional Chinese copy: QuestXXX_最终版_含中文备注.pptx

Avoid hand-zipping PPTX OpenXML unless there is no better option.

6. Add Chinese Speaker Notes

Create a UTF-8 source notes file first:

SLIDE_NOTES_ZH.md

Use one numbered section per slide. Notes should explain:

• What the audience sees.
• What this slide is meant to say.
• How it connects to the experiment's method/result logic.
• Any caveats, failures, or interpretation boundaries.

Then insert notes into the PPTX.

On Windows, use PowerPoint COM automation when available:

• Create/open the PPTX in PowerPoint.
• Add each image full-slide.
• Insert notes text through the Notes Page text placeholders.
• Save as .pptx.
• Verify the file opens without repair prompts if possible.

After creation, verify by opening the .pptx as a zip or using PowerPoint that:

• Notes count matches slide count.
• Chinese text remains readable, not ????.
• Images are present in all slides.

7. Report Outputs

At the end, give the user:

• Exact output folder.
• Main PPTX path.
• Notes source path.
• Prompt file path.
• Image count and naming pattern.
• Any limitation, such as not being able to open PowerPoint GUI in the environment.

Use clickable absolute file links when responding from Codex.

Practical Defaults

• Default language for notes: Chinese.
• Default slide format: widescreen 16:9.
• Default slide layout: image fills slide, no extra titles unless user asks.
• Default image count: choose enough to explain the experiment clearly; for a full research experiment, 12-20 images is typical.
• Default folder naming: <experiment>_ppt_ai_images.

Common Triggers

Example user requests that should use this skill:

• “根据当前目录的内容来做组会ppt。”
• “做组会PPT，图片和备注也一起做。”
• “读取这个实验目录，生成多张图片并制作 PPT+备注。”
• “把当前实验做成一份答辩 PPT。”
• “根据结果 export 生成图片提示词，再做 PPT。”
• “以后按上次实验三的流程来。”
• “为这个实验目录生成 AI 图和中文讲稿备注。”