pptx2md

pptx2md-go

Drop a .pptx, get clean Markdown. Pure Go — no pandoc, no Python, no runtime deps.

Why pptx2md?

Every .pptx-to-Markdown converter I could find needed Python. This one doesn't — it's pure Go, a single static binary, built straight from the OOXML spec (the gory details below).

Other converters	pptx2md
Python runtime + a dependency tree	Single static binary, zero runtime deps
pandoc can't even read .pptx	Pure-Go OOXML parser, in-process
Markdown styled for humans	Semantic Markdown aimed at AI agents
pip install roulette	One checksum-verified download

What it extracts

• Slide titles → ## Slide N: Title
• Bullets (nested by indent level) and paragraphs
• Tables → Markdown tables
• Images → [IMAGE: alt] text placeholders (no image bytes; output targets AI agents)
• Speaker notes → > **Notes:** …
• Slides in true presentation order, separated by ---

Install

As a Claude Code skill

/plugin marketplace add greenstevester/pptx2md-go
/plugin install pptx2md@pptx2md

Restart Claude Code, then ask: "What pptx skills do you have?" (Adding the marketplace only registers it — the second command installs the plugin.) The conversion binary is downloaded (and checksum-verified) for your platform on first use.

As a standalone binary

curl -fsSL https://raw.githubusercontent.com/greenstevester/pptx2md-go/main/skill/install.sh | bash

Detects your OS/arch, pulls the matching build from the latest release, and verifies its sha256 — or just grab a tarball/zip from the releases page yourself.

Usage

  ┌─────────┐     ┌──────────────┐     ┌─────────────┐
  │  .pptx  │ ──▶ │  pptx-to-md  │ ──▶ │  clean .md  │
  └─────────┘     └──────────────┘     └─────────────┘
                  extract → render → postprocess

pptx-to-md deck.pptx                # writes deck.md
pptx-to-md deck.pptx out.md         # writes out.md
pptx-to-md deck.pptx --stdout       # writes to stdout
pptx-to-md postprocess deck.md      # re-run the cleanup pass on existing Markdown

Build & test

make build    # -> build/pptx-to-md
make test     # go test -race ./...
make ci       # full local CI pipeline

The engine lives in internal/pptx (package pptx); the thin CLI entrypoint is cmd/pptx-to-md.

Releases

Push to main auto-tags (BREAKING→major, feat→minor, fix→patch) and runs GoReleaser to publish platform binaries (linux/macos/windows × amd64/arm64) + checksums, consumed by the skill installer.

Background

Sibling of word-doc-to-md-skill-go — same "output for AI agents, not humans" philosophy, different engine: pandoc has a .pptx writer but can't read one, so conversion is done in-process in pure Go.

The .pptx format is fully specified — the catch is that the spec is enormous. It's part of the Office Open XML (OOXML) standard that also defines Word and Excel. Canonical references:

• ECMA-376 — the Office Open XML standard, with the .xsd schemas bundled as electronic inserts (ecma-international.org)
• ISO/IEC 29500 — the ISO/IEC ratification of the same standard (iso.org catalogue; the current edition is free via ISO ITTF)
• Open XML SDK documentation (learn.microsoft.com)
• The complete XML schemas (.xsd) for OOXML ship as normative inserts to the standard (and with the Open XML SDK); a browsable copy lives at schemas.liquid-technologies.com

ECMA-376 alone runs to several thousand pages across four parts: (1) fundamentals & packaging, (2) Open Packaging Conventions — the ZIP container, (3) markup language reference, and (4) transitional migration features.

Under the hood a .pptx is just a ZIP of XML documents linked through relationship (.rels) files: