Visual explainer skill

Visual Explainer

Turn what you know into something a person understands in ten seconds.

words  ──►  diagram  ──►  interactive page  ──►  animated explainer
 slow        better        DEFAULT HERE           when asked

Climb as high as the request allows. The default output is one HTML page whose spine is figures. Text is caption, not content.

Deliver

• Write ~/.agent/diagrams/<descriptive-name>.html, or the path the user gives. One complete file: inline CSS, JS, and SVG favicon. CDN only for fonts and libraries.
• Pi: visual_explainer prepare → render (ask before prepare unless a visual was requested). MCP: render tools default to open:false. Elsewhere: write the file and open it. Use viewer:"glimpse" only on request.
• If a terminal table would have 4+ rows or 3+ columns, render HTML and reply with one summary line.
• Write a Markdown companion (<name>.md beside the HTML) only when the user asks for AI-readable output or a source brief. HTML stays the source. Ask before you overwrite one.
• Quick mode: only for a literal --quick on /generate-web-diagram, /diff-review, /plan-review, /project-recap. Do the same research, read ./quick/README.md and ./quick/schema.json, emit the JSON spec, and render with action:"render_quick" (Pi) or node <this skill's directory>/quick/render.mjs spec.json out.html (resolve the path from the skill directory, not the user's repo). If the content does not fit or rendering fails, use full HTML.

Show, don't tell

1. One claim per figure. Put each figure in <figure>; the <figcaption> states the claim in one sentence.
2. Figures lead, words label. Most sections open with a figure. Reasoning, definitions, trade-offs, and open questions can lead with text, then earn a figure when one helps. Keep the lead to a sentence or two, and prose outside captions light, roughly a short paragraph per screen. Write more when the reader needs the reasoning, not to restate the picture. If a sentence describes the picture, make it a label in the picture.
3. First viewport = the answer. The main idea as a picture plus one sentence. To teach a concept, it can be the question and the picture that frames it, with the answer one scroll away. No decorative hero.
4. Draw the mechanism, not the name. A request path through a cache beats a box labeled "cache". Label every arrow with a verb: writes, invalidates, polls 30s.
5. Draw the difference. To compare options, show the edge or box each one adds or removes.
6. Encode state in form. Shape, position, and pattern, plus color. Never color alone.
7. Every number gets a picture. Waffle, bars, or sparkline beside it. A number inside a sentence is lost.
8. Cases become small multiples. Failure modes, options, environments: the same mini diagram once per case, not a table of sentences.

Shape the page to the content. These are starting points, not templates: merge, reorder, drop, or add sections, and skip any that would be empty.

Page	Typical shape
Concept explainer	question → intuition picture → mechanism (stepper) → edge cases → what to remember
Visual plan	target state → current vs. proposed → sequence → risks → done when
Review or audit	verdict → evidence figures → risks → next steps
Comparison	the difference drawn side by side → trade-offs → recommendation
Essay or long read	Paper register: text column with figures beside or between it; more prose is fine
One diagram	a single large figure, a headline, and a caption; no sections

Content	Figure
Architecture, data flow, pipeline, state, sequence, schema, before/after	Hand-drawn inline SVG → references/diagrams.md
Cards, timelines, file maps, side-by-side	CSS grid/flex
Scenarios, failure modes, options	Small multiples → references/diagrams.md
Matrix, audit, many rows of data	<table> with status chips
Rates, shares, metrics, trends	Waffle, bars, sparklines in SVG; Chart.js only for many interactive series
Depth that carries data: embeddings, spatial layouts, geometry	three.js → references/diagrams.md
A process that changes over time	Stepper or scene player → references/diagrams.md
A setting the reader should feel: TTL, rollout %, a policy	Live figure → references/diagrams.md
Slide deck	references/slides.md + templates/slide-deck.html

Draw every diagram by hand. Hand-drawn SVG gives exact placement, page fonts, theme tokens, and animation. Use Mermaid only when the user asks for it or supplies Mermaid source (see Known traps). templates/page.html is the reference build. Copy its parts (tokens, kit CSS, scripts, components), not its outline, and swap in the register the content needs.

Words

Write about 80% of the way to ASD-STE100 (Simplified Technical English):

• Answer first, detail after. Headings state the takeaway ("Cache hits skip Postgres"), not the topic ("Caching").
• One idea per sentence. Mostly short sentences (around 20 words or under). Active voice, present tense.
• Short paragraphs, usually 1–3 sentences. Bold the one key phrase, if any.
• One term per concept, the same every time. Name things by what the reader sees, not by internal structure.
• No idioms, metaphors, filler, or hedges. Specific beats clever. Controls say exactly what they do.
• Use numbered lists for steps. Use plain words over jargon. Spell out an abbreviation the first time.

Look

Aim for the standard of the best research and engineering pages: exact, calm, visual, and specific to the subject. The reader should feel that someone who understands the system drew every line on purpose, for this topic and no other.

Always read references/style-guide.md before HTML. Precedence: the user's words → the project's design system → the style guide.

content ──► register ──────────► subject ──► topic motif (2–3 touches)
reviews, metrics   Instrument    Geist · near-black · amber        accent hue from the subject
architecture       Blueprint     IBM Plex · blue-black grid        stage texture · glyphs
concepts, reading  Paper         Newsreader + Atkinson · ink blue  one type habit of the domain
recaps, decks      Editorial     Instrument Serif + Sans           the domain's diagram convention

• Set it like a good magazine. A 12-column grid with one flush-left axis. Few type sizes with big jumps. Real typographic characters. Charts labeled directly, each with one annotation. The style guide's Typography, Layout, and Figures sections hold the rules.
• Quiet ground, one signal. Neutrals carry the page. The accent marks only what the reader must look at now.
• Craft is part of the meaning. Raised nodes on a dot-grid stage, shapes that say what a thing is, a halo on the focal element, a glow on the hot path, the number leading the headline. Each effect points at something.
• The figure is the interface. Prose terms and figure elements light up together. Overview first, detail on demand.
• Motion shows change or flow. Blocks rise in once, numbers count up, dots move along edges at the real rate. No parallax, scroll-jacking, or decorative loops. Under reduced motion, show the final state.
• Stay on the scale. No sizes, gaps, radii, or colors outside the style guide.
• Two schemes: tokens on :root, and prefers-color-scheme redefines tokens only.
• Never: Inter, Roboto, Arial, or system-ui as the only font; violet or fuchsia Tailwind accents; neon; glassmorphism; gradient text, backgrounds, or blobs; clip-art, mascots, or emoji markers; centered everything; an accent bar on a rounded card; 01/02/03 when order does not matter.
• Reading comfort: html{font-size:17px}, rem elsewhere. Body ≥ 16px, labels ≥ 12px, SVG labels ≥ 12px as rendered. 45–68ch, left-aligned. Text ≥ 4.5:1, lines ≥ 3:1. Italics and uppercase only for a few words. One focal point per viewport. Show position (2 / 5, section nav). Slides keep their clamp() px scale.
• Themes: switchable themes or fonts, or a named palette (Dracula, Nord…) → references/themes.md.

Known traps

• Set min-width:0 on grid/flex children, grid-template-columns:minmax(0,1fr) on single-column grids, and overflow-wrap:anywhere on paths. Put wide tables, code, and SVG in a scroll container.
• Do not put display:flex on <li> when its markers matter.
• If the user asks for Mermaid: theme:'base' with themeVariables read from getComputedStyle (Mermaid cannot read CSS variables), re-rendered on a theme or font switch. Quote labels that hold punctuation; <br/>, not \n. Natural size in a scroll box, never shrunk to fit. No page class named .node, .label, .nodeLabel, .edgeLabel, .cluster, .marker, .note, .actor, or .commit; Mermaid uses them. For zoom and pan, copy templates/mermaid-flowchart.html from Older recipes.
• Wrap history.replaceState in try/catch. It throws on file:// pages.
• Add section navigation (sticky TOC with scroll-spy) only for 4+ sections.
• Respect prefers-reduced-motion: copy the template's .js-motion pattern, so the final state shows without JS, in print, and under reduced motion.

Animate

When the user asks for an animated explainer or a video:

• Default: an HTML scene player. SVG scenes with play, pause, scrub, and captions, all in one file. See references/diagrams.md.
• Real video (3Blue1Brown style): first check which tools are installed. Use Manim, Remotion, or Motion Canvas for visuals and ffmpeg to mux. For narration, use ElevenLabs if the user gives a key. If not, use local TTS (macOS say, Piper, Kokoro). Do not install tools or spend API credit without asking.
• Script first. One claim per scene, a sentence or two of narration, and the picture changes with every sentence.

Slides and PPTX

Make slides only when asked (/generate-slides, --slides). Read references/slides.md. Export PPTX only on request or with --pptx: build the HTML deck first, then run visual-explainer-pptx deck.html deck.pptx (or node ./pptx/export.mjs from a checkout). Tell the user that HTML stays the source of truth. The PPTX has no animation, navigation, responsive layout, custom fonts, live diagrams, or JS.

Images

Optional. If surf or another image tool is available, you can embed generated images as base64 for a hero or concept art. Never use images for data or structure. The page must work without them.

Older recipes (optional)

Longer how-to files from v0.11.0, outside this skill. Fetch one only when you need its mechanics: https://raw.githubusercontent.com/nicobailon/visual-explainer/v0.11.0/plugins/visual-explainer/ + references/css-patterns.md (layout and CSS animation recipes) · references/libraries.md (Chart.js, anime.js, font pairs) · references/responsive-nav.md (sticky TOC with scroll-spy) · references/slide-patterns.md (slide layouts, transitions) · templates/architecture.html · templates/data-table.html · templates/mermaid-flowchart.html. They predate this guide: take the mechanics, keep this guide's look and rules.

Before delivery

□ one complete HTML file at the path; opens with no console errors
□ first viewport: main idea as a picture + one sentence
□ figures outnumber prose paragraphs; the lead and prose are short enough that the pictures carry the page
□ every number has a picture; scenarios are small multiples, not a sentence table
□ topic motif: 2–3 touches taken from the subject; page still reads without them
□ each figure: <figure> + "Fig. N" claim figcaption; role="img" + aria-label on the drawing
□ key terms linked with data-ref where hovering helps
□ no horizontal overflow at 1280px or 390px wide
□ both color schemes work (or one theme was deliberate)
□ type in rem; body ≥16px, labels ≥12px; all text ≥4.5:1 in both schemes; visible keyboard focus
□ headings state takeaways; paragraphs stay short
□ every size, gap, and radius is on the style-guide scale; polish pass done
□ typography: real dashes, minus, ×, curly quotes; no-break space before units; tabular figures in data
□ layout: one flush-left axis, consistent section openers, varied rhythm; blurred, each screen still shows one dominant element
□ register fonts and neutrals used exactly; would not pass for a generic dark/violet template
□ slides: each fits, nav chrome works, all source items covered, delivery check passes