菜单
Extract a structured evidence bundle from a .pptx deck, then write a real grounded.md from the bundle.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Unpack a ZIP archive, inventory its files, run the corresponding child grounding skill for each supported child file, and then write a real archive-level grounded.md.
Convert a raw document into a structured grounding note for downstream research and summarization.
Run focused literature and web research from a grounded note. Use when a grounded note already exists and you want targeted research results, opened-link evidence, deeper per-paper analysis materials, optional downloaded literature, and a two-stage literature output (`lit_initial.md` then refined `lit.md`).
Review a research report draft with a structured scoring rubric, run a bounded repair loop when needed, and produce the final deliverable report.
Create a rich, evidence-preserving research report draft from a grounded note and its follow-up literature result. This is the main report-writing stage of the middle pipeline, not a compression memo.
Use the input path to select the correct downstream grounding pipeline and continue execution until the selected grounding workflow is completed.
| name | pptx-grounding |
| description | Extract a structured evidence bundle from a .pptx deck, then write a real grounded.md from the bundle. |
| allowed-tools | Bash, Read, Write, Edit, Grep, Glob |
This skill handles PowerPoint (.pptx) inputs in the any-input -> grounding -> downstream research / summary / report pipeline.
Its job is not to write a polished final report. Its job is to produce a stable, structured, reusable evidence bundle for a deck. After that, the agent must read the bundle and write a real grounded.md.
Use this skill for common PowerPoint decks such as:
.pptx where the main evidence lives in slide titles, bullets, text boxes, images, tables, charts, and speaker notesUse this skill when slide order, per-slide structure, and per-slide assets matter.
This skill is not the same as document-grounding:
This skill is also not table-grounding:
This skill should extract and organize the following when present:
AssetRef insertion into extracted.mdThis skill does not aim to handle the following:
When using this skill, you must follow this workflow:
First run the existing script entrypoint:
bash .cursor/skills/pptx-grounding/scripts/run.sh <input_pptx> <output_root>
The script generates an evidence bundle only.
It does not generate grounded.md.
After the bundle is generated, read:
extracted.mdextracted_meta.jsonslide_index.jsonasset_index.jsonIf AssetRef blocks are present in extracted.md, inspect the referenced assets before writing grounded.md.
This includes, when present:
assets/images/*assets/tables/*assets/charts/*Then write a real grounded.md into the same bundle directory.
Do not create a placeholder grounded.md.
If grounded.md exists, it must be a real grounding note written after inspecting the generated bundle.
The task is not complete if only the extraction bundle exists.
The task is also not complete if grounded.md is written without first generating and using the bundle evidence.
bash .cursor/skills/pptx-grounding/scripts/run.sh <input_pptx> <output_root>
input_pptx: path to a .pptx fileoutput_root: parent directory under which the bundle should be createdExample:
bash .cursor/skills/pptx-grounding/scripts/run.sh \
/path/to/deck.pptx \
data/grounded_notes
This will create something like:
<data/grounded_notes>/<ground_id>/
The script should create:
<pptx_bundle>/
├─ extracted.md
├─ extracted_meta.json
├─ slide_index.json
├─ asset_index.json
├─ assets/
│ ├─ images/
│ ├─ tables/
│ └─ charts/
└─ grounded.md # written later by the agent, not by the script
extracted.mdHuman-readable entrypoint for the agent.
It should include:
AssetRef blocks for images / tables / chartsextracted_meta.jsonGlobal deck metadata, such as:
slide_index.jsonMachine-readable slide-level index.
Each slide record should include:
asset_index.jsonGlobal asset registry for all exported assets.
Each asset record should include:
Extract separately when possible.
Preserve paragraph order and indentation level where possible.
Extract non-title text separately from bullet blocks.
Extract notes text when present.
Export embedded images into assets/images/ and register them in asset_index.json.
For native PowerPoint tables:
For native PowerPoint charts, do lightweight best-effort export:
If chart extraction is partial, record that clearly in metadata rather than pretending extraction succeeded fully.
After reading the bundle, the agent should write a stable grounding note such as:
# PPTX Grounding
## 1. Main Topic / Purpose
## 2. Deck Structure / Narrative Flow
## 3. Main Points
## 4. Important Evidence and Assets
## 5. Speaker Notes Signals (only if present and materially useful)
## 6. Gaps / Risks / Ambiguities
## 7. Suggested Next Steps
## 8. Search Keywords
grounded.md should be a stable intermediate grounding note, not a polished final report.
It should:
It should not:
Do not write that charts / tables / images / speaker notes are "missing" unless there is positive evidence that the deck contains them and extraction failed.
If an asset type is simply not present in the PPTX, do one of the following:
grounded.md, orExamples of acceptable neutral wording:
No speaker notes were detected in this deck.No native PowerPoint charts were detected in this deck.Do not present these as risks, failures, or missing assets by default.
If extraction is partial or some assets fail to export:
asset_index.json or metadataIf a deck is too complex for full extraction, produce the best available bundle and make the limitations explicit.