napkin
Skill1 starsChanged 40 days agoMITView on GitHub
Explicit only — type /napkin <topic>. Explains a topic as a picture with few words: an HTML artifact by default, or a standalone SVG/PNG when you ask for one to embed. Offers candidate shapes to pick from before drawing.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- letitbk/claude-academic-setuptheir own version, first seen Nov 1, 2026
- github/awesome-copilottheir own version, first seen Oct 25, 2026
- 01sadra/napkintheir own version, first seen Aug 20, 2026
- ReaperOAK/ForgeOStheir own version, first seen Jun 8, 2026
- madeinlowcode/skills-claude-codetheir own version, first seen Apr 3, 2026
- seaneoliver/claude-code-startertheir own version, first seen Mar 17, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- github/awesome-copilot★ 41ksame name
- blader/napkin★ 591same name
- letitbk/claude-academic-setup★ 48same name
- madeinlowcode/skills-claude-code★ 7same name
- 01sadra/napkin★ 2same name
- seaneoliver/claude-code-starter★ 2same name
- ReaperOAK/ForgeOS★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- harness-auditSkill
- skill-benchSkill
- ask-biginSkill
- bigin-harness-setupSkill
- contract-syncSkill
- debug-workflowSkill
- discovery-workflowSkill
- epic-workflowSkill
- flutter-figma-handoffSkill
- go-scaffoldSkill
- knowledge-distillSkill
- model-routerSkill
- next-scaffoldSkill
- nodejs-scaffoldSkill
- nuxt-marketing-scaffoldSkill
- nuxt-scaffoldSkill
- nuxt-ui-figma-handoffSkill
- project-scaffoldSkill
- session-handoffSkill
- sprint-distillSkill
- task-workflowSkill
- write-testsSkill
The file
We split this skill into the rules it gives an agent
9 rules. 0 are common, 6 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
6
Distinctive
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 31
Default to the artifact. Choose SVG or PNG only when they say where it's going — "for the handbook", "to paste in Slack", "embed this".
- Very specific: line 59
Offer 2–4, then draw one — inline: Ask with AskUserQuestion, header Shape, one option per candidate.
- Distinctive: line 64
Do not write a chooser file.
- Very specific: line 68
Put the *claim* in each option's description, not just the shape's name — "asserts one thing follows another.
- Distinctive: line 78
Only one shape is honest. Steps with a decision in the middle are a flow.
- Distinctive: line 100
A third outcome usually needs a second decision node: If the third branch turns on a *different question*, a third arrow off one diamond …
- Very specific: line 105
The
aria-labelis the diagram in prose: Every exit, including ones added later. - Distinctive: line 119
Run it on what you produced, and on the file you embedded into.
- Distinctive: line 125
A screenshot would be truer: Don't draw a UI that already exists.
“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
napkin
Explain like the reader knows nothing about this topic — big picture, few words.
A napkin sketch: the simplest drawing that makes something clear.
Adapted from the MIT-licensed eli5 plugin v1.0.0 by Thariq Shihipar, installed from the
claude-community marketplace; no repository URL is published with it, so none is linked here.
What this version adds: an embeddable SVG/PNG output, and the diagram rules below — each of
which is here because it shipped as a visible defect first.
Topic: $ARGUMENTS
Pick the output first
| They want | Give them |
| To understand something (default) | HTML artifact — big visuals, few words, one idea per screen |
| A diagram for a doc, handbook, or README | Standalone SVG |
| A diagram for a slide, an issue, or chat | PNG, converted from that SVG |
Default to the artifact. Choose SVG or PNG only when they say where it's going — "for the handbook", "to paste in Slack", "embed this". If the target is a page you can read, match its existing diagrams: same CSS custom properties, same arrow marker, same type scale. A diagram that doesn't match its neighbours reads as imported.
PNG needs a converter. Probe in order and use the first that exists:
rsvg-convert -w 1600 in.svg -o out.png · magick -density 200 in.svg out.png ·
inkscape in.svg -o out.png -w 1600 · resvg in.svg out.png · on macOS,
qlmanage -t -s 1600 -o . in.svg (writes in.svg.png). If none is installed, deliver the SVG,
say which converters would work, and don't pretend a PNG was produced.
Then pick the shape — offer it, don't assume it
A shape is a claim. A flow says this happens after that; a comparison says these differ along the same axes; a timeline says the distance between them means something. Pick the wrong one and the diagram is confidently wrong before a single label is written — the one defect careful geometry never catches.
| Shape | What it asserts | Wrong when |
| Flow | One thing happens after another; every branch is a decision | The order is arbitrary |
| Comparison | Two or more things differ along the same axes | An axis doesn't apply to both |
| Timeline | Order is calendar order, and the gaps carry meaning | You don't know the dates |
| Hierarchy | Children belong to parents — containment or reporting | The links are peer-to-peer |
| Cycle | The last step feeds the first, and it runs again | It terminates |
| Layers | Each level rests on the one below and talks only to its neighbours | Anything skips a level |
| Anatomy | One thing, called out in parts | There is more than one thing |
Offer 2–4, then draw one — inline. Ask with AskUserQuestion, header Shape, one option per
candidate. Each option's preview is an ASCII sketch of the structure: eight marks at most,
placeholder labels of a word or two, no real content. They render side by side in the terminal and
answer in a keystroke.
Do not write a chooser file. A picture that costs a file open and a window switch is slower than the question it was asking, and it leaves a stray artifact behind for a decision that took two seconds.
Put the claim in each option's description, not just the shape's name — "asserts one thing
follows another; answers how a task gets routed". The name says what it will look like; the
description says what it would commit you to, which is the thing actually being chosen.
In the message above the question, name the shapes you rejected and why, one line each. That is where the table does its visible work, and it is often what changes the pick.
Skip the chooser — and say you skipped it — when:
- They named the shape: "a timeline of", "flowchart of", "compare X and Y".
- Only one shape is honest. Steps with a decision in the middle are a flow; there is no second reading to offer, and inventing one to look thorough spends a turn on nothing.
- They asked for a change to a diagram that already exists.
The answer usually arrives as a pick plus an edit; the free-text Other option carries it — "B, but colour the reject path". Take both. Then draw that one at full fidelity, and run the checker on it.
Diagram rules
SVG <text> does not wrap. Nearly everything below follows from that one fact.
- One line per
<text>. A label too long for its box runs past both rounded corners. Split it across two<text>elements and grow the box height. - Never widen one box to fit its label. A column of steps reads as a column only if the boxes line up; widening one trades a text bug for an alignment bug. Wrap the label, or reword it — the shorter wording is often the more precise one.
- Budget the label before you write it. Roughly
chars × font-size × 0.5px, againstbox width − 16. Atfont-size: 13in a 300px box that's ~43 characters; at10.5, ~54. - Compute every edge endpoint from the shape it touches. An edge meeting a box side belongs
at
y + height/2, not at a number that looked right — that number stops being the centre the moment a height changes, and nothing tells you. - A third outcome usually needs a second decision node. If the third branch turns on a different question, a third arrow off one diamond states something false. Two nodes in sequence, each with its own question, is the honest shape.
- Theme tokens, not hex. In a page that has them, use its custom properties so the diagram
follows light/dark. Standalone, define them once on the root and use
prefers-color-scheme. - The
aria-labelis the diagram in prose. Every exit, including ones added later. It is the only version a screen reader gets, and the first thing to go stale.
Check it before you hand it over
node ${CLAUDE_SKILL_DIR}/scripts/check_diagram.mjs path/to/file.svg # or the .html it lives inReports label overflow, mixed widths in a centred column, overlapping boxes, anything outside the
viewBox, and edges that miss a face's centre. Exit 1 on any finding. It skips <svg> markup that
is really a JavaScript string, and it estimates label widths — a near-miss is worth looking at
rather than trusting in either direction.
Run it on what you produced, and on the file you embedded into. Both of the defects that motivated this skill passed a visual read and failed this check.
When not to use
- A data chart — bars, lines, pie, a plotted dataset. That's a charting job, not an explainer.
- A screenshot would be truer. Don't draw a UI that already exists; capture it.
- The answer is one sentence. A picture of one sentence is a slower sentence.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

