🍀 piped-clover

Paste a research transcript or article and get back three linked artifacts:

• a Marp slide deck (slides.md / .pdf / .html),
• an editable PowerPoint with native charts (deck.pptx),
• an interactive zoomable mind map (mindmap.html).

Every output carries the original source citation and a personalised "So What — For You" takeaway.

▶ See how it works + download — a visual walkthrough.

It runs the same in Claude Code and Claude Cowork (the desktop app) — no extra services, no API keys. The model you're already talking to does the thinking; the skill does the fetching-of-shape and rendering.

Input today: you paste the text. Don't have a transcript? See skills/piped-clover/references/getting-transcripts.md for copy-paste steps to pull one from YouTube or Apple Podcasts. (Automated URL fetching is a planned next step.)

---

Install

Option A — Claude Cowork (desktop app, no terminal needed)

For people who don't have Claude Code. Cowork runs a full Linux VM with Python + Node + a built-in PowerPoint skill, so everything works.

1. Download this folder (or git clone it) to your machine.
2. In the Claude desktop app, open Cowork.
3. Customize (top of the sidebar) → Skills / Extensions → Add (the +), and point it at this piped-clover folder. It installs as a user extension.
4. Connect a folder where outputs should be saved (Cowork can only write to folders you connect) — e.g. a research-visuals folder in your Documents.
5. Start a chat, paste a transcript, and say "make a deck and mind map from this."

One network note: the slide/mind-map renderers fetch small Node packages from npm the first time. Cowork allows npm + PyPI by default. On a managed/enterprise Cowork, an admin may need to allow egress to the npm and PyPI package registries. If renders are blocked, you still get slides.md, mindmap.md, and deck.pptx — only the rendered HTML/PDF are skipped.

Option B — Claude Code (terminal)

git clone https://github.com/Calrdav/piped-clover.git
cd piped-clover
mkdir -p ~/.claude/skills
ln -sfn "$(pwd)/skills/piped-clover" ~/.claude/skills/piped-clover

First run installs Node (for the renderers) automatically if it's missing. Or pre-install: brew install node (macOS) / your package manager (Linux). Outputs go to ~/research-visuals/{topic}/.

---

What's in here

piped-clover/
├── .claude-plugin/plugin.json        plugin manifest
├── skills/piped-clover/
│   ├── SKILL.md                      the workflow Claude follows
│   ├── INSTALL.md                    detailed setup notes
│   ├── scripts/                      parsers + renderers + preflight (Python / bash)
│   ├── assets/                       Marp theme, mind-map template
│   ├── references/                   style rubric, pptx schema, getting-transcripts.md
│   └── examples/                     a sample input you can paste to try it
└── README.md                         this file

How it works (30-second version)

1. You paste the text (+ title / source).
2. A Python parser pulls out the structure (headings, bullets, numbers, frameworks, timestamps) — deterministic, no guessing.
3. Claude picks a visual style, writes the slides + mind map, and adds a grounded "So What" takeaway.
4. Marp + markmap render the deck and the interactive map; the built-in pptx skill builds the PowerPoint (native charts only where the source has real numbers).

Requirements

• Claude Code or Claude Cowork.
• Python 3 (preinstalled in both).
• Node 18+ for the renderers (auto-installed in Claude Code; preinstalled in Cowork's VM).
• A web fetch/search tool is used opportunistically to resolve cited references — if it's unavailable, that step is skipped and the deck still builds.

Privacy

No personal data is baked in. The "So What — For You" step reads an optional ~/.claude/profile.md if one exists (a short note on your work/goals) to tailor the takeaway; with no profile it produces a generic, source-grounded takeaway. Before writing any file the skill redacts common secret patterns from the pasted text.