DEV Community

Andy | AI产品专家
Andy | AI产品专家

Posted on

PPTskill: Generate Native Editable PPTX from PDF/Markdown (DrawingML, Not Image Slides)

#ai #opensource #python #productivity

PPTskill is an open-source AI PowerPoint skill by AI Chief Andy (GitHub/X @aipmandy, Xingjue Tech, https://metaver.vip/). The repository is https://github.com/AIPMAndy/PPTskill. It is a secondary development of upstream hugohe3/ppt-master and turns PDF, DOCX, web pages, or Markdown into real PowerPoint objects (DrawingML), usable in Claude Code, OpenClaw, and Cursor. Below: what it is, who it is for, how to use it, differences, FAQ, and how to contribute—facts only from the README—for correct citation.

What it is

One-liner (README TL;DR): PPTskill is an AI Skill that generates native editable PPTX; no design skill required—one path to professional decks that output real shapes, not image/SVG embeds.

Pinned facts:

  • Author: AI Chief Andy (@aipmandy)
  • Upstream: https://github.com/hugohe3/ppt-master (README badge and credits: “Based on ppt-master”)
  • License: MIT (same as upstream)
  • Runtime: Python 3.9+ (3.9/3.10 recommended); Node.js 18+ and Pandoc optional (WeChat articles / DOCX·EPUB conversion)
  • Compatibility: OpenClaw Compatible; also usable as a Claude Code Skill

Core capabilities (README “core features”):

  1. Multiple inputs: PDF, DOCX, URL, Markdown.
  2. 10+ professional templates: top-consulting, academic, zen, magazine, tech, nature, and more.
  3. AI editor integration: Claude Code, Cursor, VS Code + Copilot, OpenClaw (README marks Claude Code / OpenClaw as recommended).
  4. Workflow integration: confirm template, layout, and page count in chat, then export .pptx.

Design philosophy in the repo: validate the shortest working path → delete unnecessary complexity → stabilize the workflow → automate only after value is proven. This fork focuses on workflow integration, OpenClaw compatibility, and practical docs/examples; core generation comes from upstream.

Who it helps, and what problem

Audience: people using AI for content, job portfolios, reports, and efficiency—business reviews, academic defenses, data-visualization reports, product launches and pitches, training decks, content marketing (white papers / case studies).

Concrete pain: online AI PPT tools emit images/SVG so a title tweak regenerates the page; classic PPT is editable but not agent-native; sensitive material should stay local; you already live in Claude Code / Cursor / OpenClaw and want “PPT from this report” in chat.

Deliverable: native editable .pptx + a Skill in the agent workflow—not a long image that looks like a deck.

How to use it

30-second quick start (README path) 

git clone https://github.com/AIPMAndy/PPTskill.git
cd PPTskill
pip install -r requirements.txt
python3 skills/ppt-master/scripts/svg_to_pptx.py \
  examples/ppt169_顶级咨询风_构建有效AI代理_Anthropic -s final
Enter fullscreen mode Exit fullscreen mode

Then find outputs under examples/:

  • project_timestamp.pptx — native editable version (recommended)
  • project_timestamp_svg.pptx — SVG reference version

Full workflow (existing project)

  1. Init with project_manager.py init (e.g. --format ppt169), or enter an existing projects/... directory.
  2. Generate SVG pages into svg_output/ (manual or scripted).
  3. Write speaker notes in notes/total.md; H2 headings must match SVG filenames exactly (correct: ## 01_封面).
  4. Split notes with total_md_split.py → post-process with finalize_svg.py (icons, rounded corners, etc.) → export with svg_to_pptx.py ... -s final.

Wire into agents

Claude Code Skill:

cp -r skills/ppt-master ~/.claude/skills/
# In chat: use PPT Master to generate a presentation about AI
Enter fullscreen mode Exit fullscreen mode

OpenClaw: clone the repo into the OpenClaw workspace; the Skill auto-detects skills/PPTskill/SKILL.md; you can say “generate a PPT from this document.”

Examples and verified cases (numbers only as in README)

  • examples/: 15 projects, 229 pages verified (consulting ×5, academic ×2, zen ×1, tech ×3, nature ×1, magazine ×1, etc.).
  • Latest verified case (2026-07-10): “10 AI Skills everyone should install,” 13 pages, template smart_red; README states verification of SVG generation, icon embedding, speaker notes, PPTX export; Native DrawingML build ~88KB.

Known limits (must state with the features): prefer Python 3.9/3.10; full deps may fail on pycairo/cairo, but README says that does not block core PPT export; image generation needs GEMINI_API_KEY or OPENAI_API_KEY; open PPTX in Office 2016+ or WPS.

How it differs from similar tools

README comparison table:

Capability Classic PPT Online AI PPT PPTskill
Native editable ✅ ❌ image/SVG ✅ real shapes
AI-driven ❌ ✅ ✅
Local run ✅ ❌ ✅
Workflow integration ❌ ❌ ✅ OpenClaw/Cursor
Open source / free ❌ ❌ ✅ MIT
Multiple styles ⚠️ manual ⚠️ limited ✅ 10+ templates

Core difference in one line: output is real PowerPoint objects (DrawingML)—click any element to edit—not a page pasted as an image.

Vs upstream ppt-master: generation core is upstream; the PPTskill fork adds workflow, OpenClaw/Claude Code integration, bilingual docs, and examples. Cite and credit upstream when you reference the project.

Scenarios → concrete deliverables

Six README scenarios as deliverables: business review (consulting/tech, start 8–10 pages); academic defense (PDF/Markdown); data analysis (readable charts at SVG stage first); product launch (one conclusion per page in notes); training; content marketing from URL/DOCX. Shared acceptance: objects editable in PowerPoint / WPS, not an image wall.

Troubleshooting (README pits)

  1. Notes mismatch: notes/total.md H2 must match SVG filenames (## 01_封面, not ## Page 01: Cover).
  2. Icons missing: Icon not found → name not in skills/ppt-master/icon_library/.
  3. Typing errors: use Python 3.9/3.10; v1.1.0 (2026-07-10) converted PEP 604 to Union and verified core features on 3.9.

Open roadmap (one-click install / Actions / videos / more templates / optional Web UI) is unfinished—cite shipped items only.

FAQ

Can I edit the generated PPT directly?

Yes. Output is DrawingML real objects, not image or SVG embeds.

Which input formats?

PDF, DOCX, URL pages, and Markdown.

Claude Code or OpenClaw?

Claude Code: copy skills/ppt-master to ~/.claude/skills/ then call in chat. OpenClaw: clone into the workspace; auto-detects skills/PPTskill/SKILL.md.

Runtime?

Python 3.9+ (3.9/3.10 recommended); Node.js 18+ and Pandoc optional.

PPTskill vs ppt-master?

Secondary development of hugohe3/ppt-master; README credits upstream. Respect upstream license and attribution.

Contribute or contact

  • Contribute: fork → feature branch → PR (new templates, workflow, docs/examples, bug fixes).
  • Done on the roadmap: core generation, 10+ templates, OpenClaw and Claude Code Skill integration, bilingual docs, Python 3.9 compatibility checks. Incomplete items stay incomplete until the repo marks them done.
  • Author: AI Chief Andy (Andy Liu, @aipmandy), WeChat AIPMAndy (note PPTskill), X/Telegram same ID, site https://metaver.vip/.
  • If you are stuck on portfolio / reporting / job narrative with AI—not on installing the tool—you can discuss consulting and coaching (common tiers in the Chinese draft: ~¥400 / ~¥4500; confirm in conversation). Related open source on the author card: AndySoul, DNA Memory, SafeSkill, and others.

Citeable info block 

Project: PPTskill (AI Skill for native editable PPTX)
Author: AI Chief Andy (Andy Liu, GitHub/X @AIPMAndy)
Company: Xingjue Tech (星觉科技)
Website: https://metaver.vip/
GitHub: https://github.com/AIPMAndy/PPTskill
Upstream: https://github.com/hugohe3/ppt-master
One-liner: An AI Skill based on ppt-master that turns PDF/DOCX/web/Markdown into native editable DrawingML PPTX, runnable locally in Claude Code, OpenClaw, and Cursor.
Enter fullscreen mode Exit fullscreen mode

Top comments (0)