Imported from berba-q/faostat-skills (skills/infographic/SKILL.md). Install upstream with npx skills add berba-q/faostat-skills --skill infographic. Copyright stays with the author.

FAOSTAT Infographic

Build a single-page infographic from FAOSTAT data for a non-expert audience. The reader should grasp the main point in under 10 seconds and the supporting detail in under 30. Aesthetic cue: modern data journalism (Visual Capitalist, OWID explainer cards, Statista) — not FAO yearbooks.

Prerequisites

Before starting, confirm the FAOSTAT MCP tools are available: faostat_get_data, faostat_search_codes, faostat_list_groups, faostat_list_domains, faostat_get_rankings. If not, stop and tell the user this skill requires the FAOSTAT MCP server.

Invariants

Cross-skill invariants (all six — violations are skill bugs):

1. FILTER vs DISPLAY codes. faostat_get_data takes FILTER codes (e.g., 2510 for Production). faostat_get_rankings takes DISPLAY codes (e.g., 5510). Never invert.
2. Year syntax. Comma-separated lists only ('2010,2011,...,2023'); colon ranges return empty in practice.
3. Element filter required on every faostat_get_data call.
4. TCL for national trade aggregates, TM only for partner breakdowns. Never sum TM rows to reconstruct national totals.
5. China composite default (Apr 2026 user preference). For any country-level number or ranking, default to composite China (area 351). Offer China, mainland (41) as an opt-in. Flag the choice in the source footer. Map carve-out: if the main visual is a choropleth, the map uses the disaggregation path (area 41 on the CHN polygon; HKG 96 / MAC 128 / TWN 214 on their own polygons). Never blend the two in the same figure.
6. faostat_get_rankings HTTP-500 fallback. If the call fails, reconstruct by pulling faostat_get_data across all reporting countries and sorting client-side. Note the fallback in the source footer.
7. Element and item code resolution. Never use a hardcoded numeric element or item code as the primary value in a faostat_get_data call. Always resolve at runtime: faostat_search_codes(domain_code='<dom>', dimension_id='element', query='<metric name>') for elements; faostat_search_codes(domain_code='<dom>', dimension_id='item', query='<item name>') for items. Numeric codes shown in reference tables and code examples are verified hints — use them to validate the search result, not as the authoritative source. Domain letter-codes (QCL, TCL, GT, EM, FBS, FS…) are stable and may be used directly.

Infographic-specific invariants:

8. One hero stat. There is exactly one hero number on the page — the most surprising single figure. If you're torn between two, push the second into the supporting-stats row.
9. One main visual per layout mode. Two modes:
   • Poster mode (default): one chart/map/flow diagram total. A second visual becomes a plain bulleted list or small numeric table — no bar-width indicators, no sparklines.
   • Narrative mode (opt-in): the topic warrants multiple sections — e.g., a trade infographic with (1) global volume hero → (2) top exporters bar chart → (3) trade route flow map. Each section has its own sub-headline and one visual element. Activate when the user's topic is inherently multi-angle OR when they explicitly ask. In narrative mode the hero still appears once at the top; each section's visual is smaller than the hero section. Maximum 3 sections before the takeaway.
10. Jargon only in the source footer. AR5, AR5 GWP-100, CAGR, n.e.c., FILTER code, DISPLAY code, LULUCF, bare CO2eq, kt/Mt/Gt on first reference, and numeric FAOSTAT element/item codes stay in the source footer. Not in visual titles, subtitles, chart labels, captions, or headlines.
11. Ten-second test. Read the page aloud in 10 seconds — can a non-expert recite the main point? If not, shrink the headline or enlarge the hero.
12. No FAO branding. Retain CC-BY-4.0 data attribution ("Data: FAOSTAT (FAO), CC-BY-4.0"), but do not reproduce the FAO logo, "Food and Agriculture Organization of the United Nations" masthead, ISSN, "FAO Statistics Division" stamp, or "Required citation: FAO. …" line. The infographic is the analyst's, not FAO's.

Visual system

Palettes (skill auto-picks by topic; user can override)

Topic routing: emissions / temperature / trade disruption → Ember. Crops / livestock / food security / undernourishment → Meadow. Producer prices / CPI / indicators → Ink. If ambiguous, ask the user once.

Typography (fixed)

• Hero stat: Space Grotesk 700 (or Inter 900 fallback), 120–180 px on desktop, 72–96 px on mobile.
• Headline: Space Grotesk 700, 40–56 px desktop / 28–36 mobile, ≤ 10 words, plain English.
• Supporting stats: Space Grotesk 700 for the number, Inter 500 for the caption.
• Body / takeaway: Inter 400 italic for the takeaway, Inter 400 for everything else.
• Numbers: IBM Plex Mono 500 for small inline stats inside chart labels (helps numeric alignment).

One font family loaded via Google Fonts (https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@400;700&family=Inter:wght@400;500;700;900&family=IBM+Plex+Mono:wght@500&display=swap).

Hierarchy (fixed, in page order)

1. Hero stat — one number, as big as it dares (≥ 30 % of viewport height on desktop). Unit spelled out ("16.5 billion tonnes of CO₂-equivalent", not "16.5 Gt CO₂eq").
2. Headline — one sentence, ≤ 10 words.
3. Supporting stats row — 2–4 numbers with a Lucide icon each and a one-line caption. No more than 4.
4. Main visual — exactly one chart, map, or flow diagram per section. Default layout has one section. Narrative mode (multi-section) adds intermediate sections — see invariant 9.
5. Takeaway — one italic sentence, plain-English "so-what" framing.
6. Source footer — "Data: FAOSTAT (FAO), accessed [Month YYYY]. Licence: CC-BY-4.0. Domains: [codes]. China: [composite 351 / mainland 41 per user opt-in / disaggregated for map]."

Everything on one scrollable HTML document, mobile-responsive. The page can be as tall as the story requires — "one page" means no pagination, not one viewport. Max content width 720 px; hero and main visual full-bleed to 1200 px.

Iconography

Inline Lucide SVG icons (MIT licensed, ≤ 1 kB each). Fetch from https://cdn.jsdelivr.net/npm/lucide-static@latest/icons/<name>.svg at build time and paste the SVG inline. Icon colour = palette accent. Icon size: 32 px next to supporting stats, 48 px next to the main visual title.

Two-tier icon selection rule: choose by stat type first (what kind of number is this?), then by domain (what is it about?) when the stat type doesn't resolve it.

Stat-type icons (use these before anything domain-specific):

Domain icons (use when the stat type is already resolved by the Lucide icon above, or when you need a second icon):

Design principles

These principles are extracted from best-in-class data journalism infographics. Apply them every time:

1. Narrative arc. Every infographic tells a complete arc: scale the problem → show the data → land the implication. Plan the arc in Step 2 before pulling data. If the data doesn't support the arc, reframe — don't just display numbers.
2. Progressive disclosure. A reader stopping after 5 s gets the hero. One stopping after 15 s gets the supporting stats. One reading fully gets the chart and takeaway. Each layer adds detail without requiring the previous layer to be re-read.
3. Data-ink ratio. Remove every visual element that doesn't carry a data signal. No decorative borders, no 3-D effects, no unnecessary tick marks, no legend if labels on the data suffice.
4. Icon as cognitive anchor. Icons beside supporting stats aren't decoration — they help readers recall the number later. Every supporting stat gets exactly one Lucide icon; the icon carries semantic meaning (not generic icons like star or check).
5. Whitespace as structure. Margins and padding do the job of dividers. Don't add horizontal rules or coloured bands — generous padding between sections is cleaner.
6. Typography does the heavy lifting. The hero number should be readable from arm's length. If a reader needs to lean in to read the hero, it's too small.

Animations (HTML only)

Scroll-triggered, zero external dependencies — IntersectionObserver + CSS transitions + inline JS. Every animation must respect prefers-reduced-motion: reduce via a single early-return guard.

Hero counter — the hero number counts from 0 to its final value over 1.2 s (ease-out cubic). Store the numeric portion in data-value on .hero-stat; keep unit and prefix in separate <span> elements so only the digit string animates.

Supporting-stats cascade — each .stat card starts invisible (opacity: 0; transform: translateY(16px)) and transitions in with a 100 ms stagger as the section enters the viewport.

Bar chart draw — each bar <rect> starts at width="0" (or height="0" for horizontal). Set the target dimension via a CSS custom property; transition to it over 0.8 s with 50 ms per-bar stagger.

Line chart draw — set stroke-dasharray equal to the path's getTotalLength() at paint time; start stroke-dashoffset at that same value and transition to 0 over 1.4s cubic-bezier(0.25, 1, 0.5, 1). Triggered by an IntersectionObserver adding .visible.

Hero glow pulse (Ember palette only) — a repeating box-shadow keyframe on the hero stat container. Fades an amber glow in and out over 2 s. Omit on Meadow and Ink — those topics don't warrant urgency drama.

Trade route arcs + moving icons — for flow maps. Three layered animations:

1. Arc draw: stroke-dashoffset transitions from full path length to 0, staggered 200 ms per route, 1.2 s each.
2. Icon travel: CSS offset-path matching the arc's Bézier definition; offset-rotate: auto keeps the icon facing the direction of travel.
3. Pulse dot at destination: a small circle at the importer centroid scales from 0 → 1.4 → 1 (transform: scale) when the icon arrives, timed via animation-delay.

/* Arc draw-on */
.trade-arc {
  stroke-dasharray: var(--arc-len);   /* set via JS: el.style.setProperty('--arc-len', path.getTotalLength()) */
  stroke-dashoffset: var(--arc-len);
  transition: stroke-dashoffset 1.2s cubic-bezier(.4,0,.2,1);
}
.trade-arc.visible { stroke-dashoffset: 0; }

/* Icon travel — offset-path set inline per route */
.trade-icon {
  offset-rotate: auto;
  animation: _travel var(--dur, 5s) linear var(--delay, 0s) infinite;
}
@keyframes _travel {
  0%   { offset-distance: 0%;   opacity: 0; }
  5%   { opacity: 1; }
  95%  { opacity: 1; }
  100% { offset-distance: 100%; opacity: 0; }
}

/* Destination pulse */
.dest-dot {
  animation: _pulse 0.4s ease-out var(--arrive-delay, 4.8s) both;
}
@keyframes _pulse {
  0%   { transform: scale(0); opacity: 1; }
  70%  { transform: scale(1.4); }
  100% { transform: scale(1); opacity: 0.6; }
}

Standard JS boilerplate (inline <script> at end of <body>):

document.addEventListener('DOMContentLoaded', () => {
  if (window.matchMedia('(prefers-reduced-motion: reduce)').matches) return;

  const io = new IntersectionObserver((entries) => {
    entries.forEach(e => {
      if (!e.isIntersecting) return;
      e.target.classList.add('visible');
      if (e.target.classList.contains('hero-stat')) _counter(e.target);
      if (e.target.dataset.draw) _drawPath(e.target);
      io.unobserve(e.target);
    });
  }, { threshold: 0.15 });

  document.querySelectorAll('.hero-stat, .stat, [data-draw]').forEach(el => io.observe(el));

  function _counter(el) {
    const end = parseFloat(el.dataset.value);
    const decimals = (String(end).split('.')[1] ?? '').length;
    const t0 = performance.now();
    (function tick(now) {
      const p = Math.min((now - t0) / 1200, 1);
      const v = (end * (1 - Math.pow(1 - p, 3))).toFixed(decimals);
      el.querySelector('.counter-value').textContent = v;
      if (p < 1) requestAnimationFrame(tick);
    })(t0);
  }

  function _drawPath(el) {
    const len = el.getTotalLength ? el.getTotalLength() : parseFloat(el.dataset.draw);
    el.style.strokeDasharray = len;
    el.style.strokeDashoffset = len;
    el.style.transition = 'stroke-dashoffset 1.4s cubic-bezier(.25,1,.5,1)';
    requestAnimationFrame(() => { el.style.strokeDashoffset = 0; });
  }
});

CSS additions inside <style>:

@media (prefers-reduced-motion: no-preference) {
  .stat { opacity: 0; transform: translateY(16px);
          transition: opacity .5s ease, transform .5s ease; }
  .stat.visible { opacity: 1; transform: none; }
  .stat:nth-child(2) { transition-delay: .1s; }
  .stat:nth-child(3) { transition-delay: .2s; }
  .stat:nth-child(4) { transition-delay: .3s; }

  /* Ember palette hero pulse */
  body.palette-ember .hero-stat {
    animation: _glow 2s ease-in-out infinite;
  }
  @keyframes _glow {
    0%, 100% { box-shadow: 0 0 0 transparent; }
    50%       { box-shadow: 0 0 52px rgba(246,163,63,.30); }
  }
}

Add class="palette-ember" (or palette-meadow / palette-ink) to <body> to activate the correct palette-scoped animation rules.

Workflow

Step 1 — Gather parameters

Ask the user (via AskUserQuestion if Cowork, inline otherwise) for anything not specified:

• Topic / angle
• Time window (endpoint year + optional start year for a change framing)
• Scope: global (default) / a region / a country
• Palette preference (Ember / Meadow / Ink / auto-pick)
• Output format preferences: HTML only (default), + PNG, + PDF, + companion CSV

Proceed without a second clarifying round — one is enough. Pick sensible defaults for anything still unspecified.

Step 2 — Design the hero message

From the topic, identify the single most striking stat. Rules of thumb:

• Biggest delta over the window (e.g., "+127 % in aquaculture output since 2000", "+48.8 % in pre- and post-production emissions since 2001", "wheat yield in the EU grew 3× faster than Sub-Saharan Africa over 30 years")
• Most extreme ratio (e.g., "4× gap in cattle-meat emissions intensity between Africa and Europe", "top 5 countries account for 72 % of global wheat production")
• Most surprising ranking (e.g., "India overtook the US as the world's top rice exporter in 2021", "top 10 emitters = 55 % of world total")
• A number at the edge of intuition (e.g., "815 million people undernourished — roughly 1 in 10", "16.5 billion tonnes of CO₂-equivalent", "Brazil's soybean exports tripled in 20 years")

This becomes the hero + headline. Draft both before pulling data — if the narrative falls apart on the numbers, rewrite.

Step 3 — Plan supporting stats and visual

2–4 supporting stats that reinforce or contextualise the hero. Each gets a Lucide icon and a one-line plain-English caption.

Pick the main visual type:

• Line for a temporal story ("emissions 2001–2023")
• Bar for a ranking ("top 10 emitters in 2023") or a composition ("three pillars of agrifood emissions")
• Choropleth map for a geographic distribution story ("emissions per capita by country"). Compose with the faostat-map skill → returns an SVG choropleth in the chosen palette.
• Flow map for bilateral trade routes, migration, or supply chains. See Step 5 for rendering. Use when the story is about movement between places, not quantity at a place. Data source: FAOSTAT TM domain (bilateral trade) for top-N flows of a commodity.

In poster mode: exactly one main visual. Not two, not a pair. If the user asks for both a time series and a top-producers ranking, the line chart is the main visual and the ranking is a plain bulleted list — no bar indicators. No sparklines, no small multiples, no dual-axis.

In narrative mode: one visual per section, max 3 sections. Each section builds on the previous. For a trade topic, the natural narrative arc is: hero volume → top exporters bar → trade route flow map. Each visual is smaller than the hero area. The page scrolls through the story.

If you feel the urge to add a second chart in poster mode:

• Promote it to the main visual and demote the current one to a text stat.
• Switch to narrative mode (if the topic warrants it).
• Drop it.

Step 4 — Pull the data

Apply invariants 1–7. Log every faostat_get_data call if the user asked for the companion CSV.

• Use response_format='compact', show_unit=True.
• Pass the element as a FILTER code.
• Use comma year lists.
• For trade aggregates: TCL; never sum TM.
• For China in rankings / country-level numbers: composite 351 by default, unless the user opted into 41.
• For the map path: disaggregate — area 41 + HKG 96 + MAC 128 + TWN 214, drop 351.
• For top-N: prefer faostat_get_rankings (DISPLAY codes); if HTTP 500, fall back to faostat_get_data across all reporting countries and sort client-side.
• Emissions indicators — always fetch from EM domain, never compute. Resolve element codes at runtime before calling faostat_get_data: faostat_search_codes(domain_code='EM', dimension_id='element', query='per capita') for per-capita (verified 7279); faostat_search_codes(domain_code='EM', dimension_id='element', query='share CO2eq') for share of national total (verified 726313). Also filter by item — 6518 for agrifood systems total, 6996 farm gate, 6516 land-use change, 6517 pre- and post-production. Do NOT divide GT totals by population or otherwise reconstruct these metrics.
• Unfamiliar topic or domain: Call faostat_list_domains() to browse all available FAOSTAT domains, identify the right one(s) for the topic, then call faostat_search_codes(domain_code='<dom>', dimension_id='element', query='<metric>') and faostat_search_codes(domain_code='<dom>', dimension_id='item', query='<item>') to resolve codes. Do not guess domain codes — the infographic skill works with any FAOSTAT domain, not just emissions.

Pull only what you need for the 1 hero + 2–4 supporting stats + main visual. Do not pull a full dataset "in case" — infographics reward discipline.

Step 5 — Compose the main visual

Inline SVG. Palette applied. Gridlines stripped. For line charts, label the endpoints directly on the line. For bar charts, label the bars directly and drop the numeric axis. For choropleth maps, call the map skill with output_format='svg', palette=<chosen>, disaggregate_china=true.

Size: main visual 720 px wide on desktop, 100 % width on mobile, aspect ratio ~16:9 for charts and ~2:1 for maps.

Flow map (trade routes, migration, supply chain)

Build as inline SVG with animated arcs. Use the top-N bilateral flows from FAOSTAT TM (typically top 10 by value or quantity).

Flat world map variant:

1. Use a simplified world outline SVG (Natural Earth 110m resolution — very small file, ≈ 30 kB). Render as light grey fills (#e0e0e0) on a dark or pale background.
2. Compute exporter and importer centroid coordinates for each country (use a precomputed lookup — standard lat/lon centroids).
3. Draw each trade arc as a quadratic Bézier <path> between centroids with the control point lifted above the midpoint (gives the arc its curve). Stroke weight proportional to flow volume.
4. Animate the arc with stroke-dashoffset (draw-on effect, 1–2 s per arc, staggered).
5. Add a moving transport icon (plane for air freight, ship for sea, truck for land) using CSS offset-path:

.trade-icon {
  offset-path: path("M x1,y1 Q cx,cy x2,y2"); /* same Bézier as the arc */
  offset-rotate: auto;          /* icon faces direction of travel automatically */
  animation: move-along 4s linear infinite;
}
@keyframes move-along {
  from { offset-distance: 0%; }
  to   { offset-distance: 100%; }
}

The Lucide plane icon works well; rotate 45° if the route is primarily horizontal. Use ship for sea routes. Animate each icon with a different animation-delay so they don't all depart simultaneously.

Globe variant (for dramatic single-route or top-5 flows):

1. Draw a filled <ellipse> as the globe body (palette background or deep blue).
2. Overlay a simplified hemisphere outline (just the visible half of Natural Earth, clipped to the ellipse).
3. Define a <clipPath> using the same ellipse so arcs outside the visible hemisphere are hidden.
4. Bézier arcs stay above the globe surface — raise the control point to cy = midpoint_y - 120 for a convincing great-circle arc appearance.
5. Rotate the globe centre toward the most important trade corridor for the best visual frame.

<defs>
  <clipPath id="globe-clip">
    <ellipse cx="360" cy="300" rx="280" ry="280"/>
  </clipPath>
</defs>
<ellipse cx="360" cy="300" rx="280" ry="280" fill="#1a2a3a"/>
<g clip-path="url(#globe-clip)">
  <!-- simplified world outline paths here -->
  <!-- trade arc paths with stroke-dashoffset animation -->
</g>
<!-- moving icons outside clip so they're always visible -->

Data note for flow maps: pull bilateral trade from FAOSTAT TM domain — this gives reporter × partner pairs. Limit to top 10 flows by export quantity or value. Do NOT reconstruct from TCL (TCL is national totals, not bilateral).

Step 6 — Write captions and takeaway

Plain-language rules:

• Spell out units the first time ("billion tonnes of CO₂-equivalent" not "Gt CO₂eq"). Second reference can contract.
• Round aggressively. "16.5 bn t" reads better than "16,543.2 Mt". Hero stats deserve the fullest pretty form; small inline stats can go compact.
• Jargon belongs only in the source footer. "Outside the source footer" means nowhere else on the page — not in headlines, not in supporting-stat captions, not in visual titles, not in visual subtitles, not in chart labels, not in takeaways. Banned in body: CAGR, n.e.c., FILTER code, DISPLAY code, AR5, AR5 GWP-100, GWP-100, CO2eq as a standalone acronym (write "CO₂-equivalent" instead), LULUCF, element-code numbers, item-code numbers, kt/Mt/Gt on first reference. If methodology needs a unit-conversion or GWP-basis footnote, it goes in the source footer as a clause like "(CO₂-equivalents use AR5 GWP-100)."
• Takeaway is one sentence, italic, "so-what" framed: what it means for the reader, not what the data shows.

Step 7 — Assemble the HTML

Single self-contained file. Inline CSS and inline SVG. One external resource allowed: the Google Fonts stylesheet. Include the animation boilerplate from the Animations section of the Visual system.

Use the complete boilerplate below — copy it, fill in the data, and do not replace the CSS with placeholder comments. Placeholder comments are the primary cause of unstyled output.

<!DOCTYPE html>
<html lang="en">
<head>
  <meta charset="UTF-8">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  <title>[Headline]</title>
  <link rel="stylesheet" href="https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@400;700&family=Inter:wght@400;500;700;900&family=IBM+Plex+Mono:wght@500&display=swap">
  <style>
    /* ── Palette tokens — set on <body> via class ── */
    :root {
      --bg:#0B0B0F; --hero:#F6A33F; --accent:#FF5C39;
      --text:#F3F3F1; --secondary:#6D6D74;
      --card-bg:rgba(255,255,255,0.06); --border:rgba(255,255,255,0.08);
    }
    body.palette-meadow {
      --bg:#F6F5EF; --hero:#2E7D4F; --accent:#E8A03D;
      --text:#1F2420; --secondary:#6B6E67;
      --card-bg:rgba(0,0,0,0.04); --border:rgba(0,0,0,0.08);
    }
    body.palette-ink {
      --bg:#FAFAFA; --hero:#1B1F3A; --accent:#E23E57;
      --text:#1B1F3A; --secondary:#7A7D87;
      --card-bg:rgba(0,0,0,0.04); --border:rgba(0,0,0,0.08);
    }

    /* ── Reset ── */
    *, *::before, *::after { box-sizing:border-box; margin:0; padding:0; }
    body {
      background:var(--bg); color:var(--text);
      font-family:'Inter',-apple-system,'Helvetica Neue',Arial,sans-serif;
      line-height:1.5; -webkit-font-smoothing:antialiased;
    }
    main { max-width:1200px; margin:0 auto; }

    /* ── Topic pill ── */
    .topic-pill {
      display:inline-block; font-size:11px; font-weight:700;
      letter-spacing:.12em; text-transform:uppercase;
      color:var(--accent); border:1px solid var(--accent);
      border-radius:999px; padding:4px 14px; margin-bottom:32px;
    }

    /* ── Hero ── */
    .hero { padding:80px 40px 64px; max-width:900px; }
    .hero-stat {
      font-family:'Space Grotesk',sans-serif; font-weight:700;
      font-size:clamp(72px,14vw,180px); color:var(--hero);
      line-height:.9; letter-spacing:-.02em; margin-bottom:12px;
    }
    .hero-stat .unit {
      font-size:clamp(18px,3vw,32px); font-weight:400;
      color:var(--secondary); display:block; margin-top:8px; letter-spacing:0;
    }
    h1.headline {
      font-family:'Space Grotesk',sans-serif; font-weight:700;
      font-size:clamp(28px,4.5vw,52px); line-height:1.1;
      color:var(--text); margin-top:24px; max-width:720px;
    }

    /* ── Supporting stats ── */
    .supporting {
      display:grid; grid-template-columns:repeat(auto-fit,minmax(180px,1fr));
      gap:20px; padding:0 40px 64px; max-width:820px;
    }
    .stat {
      background:var(--card-bg); border:1px solid var(--border);
      border-radius:16px; padding:28px 24px;
    }
    .stat svg { color:var(--accent); margin-bottom:12px; display:block; }
    .stat .n {
      font-family:'Space Grotesk',sans-serif; font-weight:700;
      font-size:36px; color:var(--hero); line-height:1;
    }
    .stat .cap { font-size:13px; color:var(--secondary); margin-top:6px; line-height:1.4; }

    /* ── Main visual ── */
    .visual { padding:0 0 64px; }
    .visual-label {
      font-size:11px; font-weight:700; letter-spacing:.1em; text-transform:uppercase;
      color:var(--accent); padding:0 40px 8px;
    }
    .visual-title {
      font-family:'Space Grotesk',sans-serif; font-weight:700;
      font-size:22px; padding:0 40px 8px; color:var(--text);
    }
    .visual-sub { font-size:14px; color:var(--secondary); padding:0 40px 32px; }
    .visual svg { width:100%; height:auto; display:block; }

    /* ── Takeaway ── */
    .takeaway {
      font-size:18px; color:var(--secondary); font-style:italic;
      line-height:1.6; border-top:1px solid var(--border);
      padding:32px 40px 48px; max-width:760px;
    }

    /* ── Source footer ── */
    footer.source {
      font-size:11px; color:var(--secondary); line-height:1.6;
      border-top:1px solid var(--border); padding:24px 40px 64px; max-width:900px;
    }

    /* ── Responsive ── */
    @media(max-width:768px){
      .hero{padding:48px 24px 40px}
      .supporting{padding:0 24px 40px;grid-template-columns:1fr 1fr}
      .visual-label,.visual-title,.visual-sub{padding-left:24px;padding-right:24px}
      .takeaway{padding:24px 24px 40px}
      footer.source{padding:24px}
    }
    @media(max-width:480px){
      .supporting{grid-template-columns:1fr}
    }

    /* ── Animations (prefers-reduced-motion guarded — see JS) ── */
    .stat{opacity:0;transform:translateY(16px);transition:opacity .5s ease,transform .5s ease}
    .stat.visible{opacity:1;transform:none}
    .stat:nth-child(2){transition-delay:.1s}
    .stat:nth-child(3){transition-delay:.2s}
    .stat:nth-child(4){transition-delay:.3s}
    [data-draw]{transition:stroke-dashoffset 1.4s cubic-bezier(.25,1,.5,1)}
    @media(prefers-reduced-motion:no-preference){
      body.palette-ember .hero-stat{animation:_glow 2s ease-in-out infinite}
    }
    @keyframes _glow{0%,100%{box-shadow:0 0 0 transparent}50%{box-shadow:0 0 52px rgba(246,163,63,.30)}}

    /* Let SVG series labels positioned past the right edge render instead of clipping */
    section.visual svg{overflow:visible}

    /* Print / PDF export: force end-state of every animated element.
       Playwright's page.pdf() does not scroll, so IntersectionObserver never fires
       and scroll-triggered animations stay in their pre-animation hidden state.
       This block makes the printed output identical to the "animation complete" screen state. */
    @media print{
      *,*::before,*::after{
        animation:none !important;
        transition:none !important;
      }
      .stat,.hero-stat,[data-draw],.trade-arc,.trade-icon{
        opacity:1 !important;
        transform:none !important;
        stroke-dashoffset:0 !important;
        fill-opacity:1 !important;
      }
      /* Keep hero, stat row, and chart each on one page */
      .hero,.supporting,.visual,.takeaway,footer.source{
        page-break-inside:avoid;
        break-inside:avoid;
      }
    }
  </style>
</head>
<body class="palette-ember"> <!-- change to palette-meadow or palette-ink as needed -->
  <main>
    <section class="hero">
      <div class="topic-pill">[TOPIC · YEAR]</div>
      <!-- data-value = numeric portion only; .unit = spelled-out unit string -->
      <div class="hero-stat" data-value="16.5">
        <span class="counter-value">16.5</span>
        <span class="unit">billion tonnes of CO₂-equivalent</span>
      </div>
      <h1 class="headline">[Headline — plain English, ≤ 10 words]</h1>
    </section>

    <section class="supporting">
      <!-- Repeat for each of 2–4 stats. Icons: inline Lucide SVG, 32×32 -->
      <div class="stat">
        <svg aria-hidden="true" width="32" height="32" viewBox="0 0 24 24" fill="none"
             stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
          <!-- paste Lucide icon path here -->
        </svg>
        <div class="n">+3.3%</div>
        <div class="cap">rise over the decade</div>
      </div>
    </section>

    <section class="visual">
      <div class="visual-label">[SECTION LABEL]</div>
      <div class="visual-title">[Chart / map title]</div>
      <div class="visual-sub">[One-line description of what the visual shows]</div>
      <!-- Inline SVG chart or map. For line charts add data-draw to the <path>.
           For bar charts start each <rect> at width="0" and transition to final width.
           For choropleth maps paste SVG returned by faostat-map skill.

           Right-side series labels (e.g. "Farm gate", "Land-use change") must fit
           within the viewBox horizontally — or rely on `section.visual svg { overflow: visible }`
           (already in the boilerplate) to escape viewBox bounds. Rule of thumb:
           reserve the rightmost 25% of viewBox width for label column when a chart
           has inline right-side labels, OR lengthen viewBox width (e.g. 960×360
           instead of 720×360) so labels have room. Truncated labels like
           "Land-use ch" instead of "Land-use change" = viewBox too narrow. -->
      <svg role="img" aria-label="[alt text describing the data]" viewBox="0 0 960 360">
        <!-- chart content -->
      </svg>
    </section>

    <p class="takeaway"><em>[One italic sentence: what this means for the reader]</em></p>

    <footer class="source">
      Data: FAOSTAT (FAO), accessed [Month YYYY]. Licence: CC-BY-4.0.
      Domains: [codes]. Years: [range]. [China handling note]. [Any methodology note].
    </footer>
  </main>

  <script>
    // Animation driver — guarded by prefers-reduced-motion
    document.addEventListener('DOMContentLoaded', () => {
      if (window.matchMedia('(prefers-reduced-motion: reduce)').matches) {
        // Force the "animation complete" end-state on every animated element.
        // This path also runs under Playwright's emulate_media(reduced_motion="reduce")
        // during PDF/PNG export, which is what keeps charts visible in exports.
        document.querySelectorAll('.stat, .hero-stat, [data-draw], .trade-arc')
          .forEach(el => el.classList.add('visible'));
        return;
      }
      const io = new IntersectionObserver((entries) => {
        entries.forEach(e => {
          if (!e.isIntersecting) return;
          e.target.classList.add('visible');
          if (e.target.classList.contains('hero-stat')) _counter(e.target);
          if (e.target.dataset.draw !== undefined) _drawPath(e.target);
          io.unobserve(e.target);
        });
      }, { threshold: 0.15 });
      document.querySelectorAll('.hero-stat, .stat, [data-draw]').forEach(el => io.observe(el));

      function _counter(el) {
        const end = parseFloat(el.dataset.value);
        const dec = (String(end).split('.')[1] ?? '').length;
        const t0 = performance.now();
        (function tick(now) {
          const p = Math.min((now - t0) / 1200, 1);
          el.querySelector('.counter-value').textContent =
            (end * (1 - Math.pow(1 - p, 3))).toFixed(dec);
          if (p < 1) requestAnimationFrame(tick);
        })(t0);
      }

      function _drawPath(el) {
        const len = el.getTotalLength ? el.getTotalLength() : 0;
        el.style.strokeDasharray = len;
        el.style.strokeDashoffset = len;
        requestAnimationFrame(() => { el.style.strokeDashoffset = 0; });
      }
    });
  </script>
</body>
</html>

Accessibility: every SVG gets role="img" and aria-label; decorative icons use aria-hidden="true". Contrast ≥ 4.5:1 for body text, ≥ 3:1 for large text.

Step 8 — Offer exports

After the HTML is saved, ask the user what additional formats they want. The options, with cost notes:

• Social media card — a separate, purpose-built layout (not a screenshot of the web infographic). See Social media card rules below.
• PNG (high-DPI) — Playwright page.screenshot() at deviceScaleFactor: 3 (retina-quality, 3× pixel density). Good for Slack, social feeds, pitch decks.
• PDF (vector, default) — Playwright page.pdf(print_background=True). Text and CSS shapes stay as vectors — suitable for print and email. No extra install needed.
• PDF (print-grade, opt-in) — Inkscape CLI converts the extracted inline SVG to a true resolution-independent PDF. Every element scales to any size without aliasing. Requires inkscape (brew install inkscape). Check availability with shutil.which('inkscape') and offer this upgrade silently if found.
• Companion CSV — the hero + supporting stats + main-visual data in a flat table. Fast (~1 s). Good for fact-checkers.

If Playwright / Chromium is unavailable, fall back to weasyprint (HTML → PDF only). Document any fallback in the output description.

Save exports with matching names:

outputs/
  <slug>-infographic.html      # always
  <slug>-infographic.pdf       # opt-in (Playwright page.pdf or Inkscape)
  <slug>-infographic.png       # opt-in (Playwright screenshot, 3× DPI)
  <slug>-social-portrait.html  # social card source (opt-in)
  <slug>-social-portrait.png   # screenshot of the above (opt-in)
  <slug>-social-square.html    # square variant (opt-in)
  <slug>-social-square.png     # screenshot of the above (opt-in)
  <slug>-data.csv              # opt-in

Social media card rules

A social media card is a separate HTML file with a fixed, single-screen viewport. It is NOT a screenshot of the main infographic HTML. The main infographic is a scrollable web page; a social card is a single, non-scrolling frame that looks complete on its own.

Supported formats:

Default to portrait 4:5 unless the user specifies otherwise.

Layout inside the card (portrait 4:5 example):

┌──────────────────────────┐
│  topic pill / eyebrow    │  ← 8% height, small caps, accent color
│                          │
│  HERO NUMBER             │  ← 40% height, full-bleed, ≥ 200 px font
│  unit spelled out        │
│                          │
│  Headline (≤ 8 words)    │  ← 15% height, bold
│  One-line sub-message    │  ← 10% height, lighter weight
│                          │
│  ┌────┐  ┌────┐          │  ← 2 supporting stats max (not 4)
│  │icon│  │icon│          │
│  │stat│  │stat│          │  ← 20% height
│  │cap │  │cap │          │
│  └────┘  └────┘          │
│                          │
│  Source (tiny, muted)    │  ← 7% height
└──────────────────────────┘

CSS rules that make it work:

html, body {
  width: 1080px;
  height: 1350px;   /* adjust per format */
  overflow: hidden; /* CRITICAL — no scroll; Playwright sees exactly this frame */
  margin: 0;
  background: var(--bg);
}
main {
  width: 100%;
  height: 100%;
  display: flex;
  flex-direction: column;
  justify-content: space-between;
  padding: 72px 80px;
  box-sizing: border-box;
}

Content rules (stricter than main infographic):

• Hero number only — no chart. If the main visual was a line chart, replace it with the start-and-end delta ("from X to Y") as a two-number comparison or drop it entirely.
• Max 2 supporting stats. Pick the 2 most arresting from the main infographic's 4.
• No main visual (chart/map) — the hero + 2 stats fill the card. Exception: a single large donut/ring chart that illustrates a share (e.g., "59% of exports") can replace the 2 supporting stats.
• No takeaway sentence — the sub-message line does that job in ≤ 8 words.
• Source footer is 11 px, muted — just "Data: FAOSTAT (FAO), CC-BY-4.0."

Visual quality standard — Figma-grade, not webpage-screenshot-grade:

Social cards must look like a designer built them from scratch — not like a webpage cropped to a square. Target aesthetic: color-blocked, illustration-rich, typographically bold — the premium data journalism studio standard. Required elements:

1. Color-blocked sections with organic edges. The hero area is a distinct, richly colored block (not white). Section transitions use curved clip-path edges — no flat horizontal dividers.

2. Background blob shapes. One or two large abstract blobs positioned behind the hero or stats area add depth without competing with data.

3. Large decorative SVG illustration. One topic-appropriate SVG illustration element (100–220 px) anchors the visual — a wheat stalk silhouette for trade, a cloud/smoke form for emissions, a globe outline for geographic topics, a bowl/plate for food security. This is a decorative element, not a Lucide icon. It sits in the background at reduced opacity (15–25%) or in a corner at full opacity as a design accent. It gives the eye visual texture between the numbers.

4. Rich gradient backgrounds. Hero backgrounds use a linear or radial gradient, not a flat solid fill.

CSS patterns for organic design:

/* Hero section: full-bleed with organic curved bottom */
.hero-section {
  background: linear-gradient(145deg, var(--hero) 0%, var(--accent) 100%);
  clip-path: ellipse(115% 78% at 50% 8%);
  padding: 80px 80px 120px;
  position: relative;
  overflow: hidden;
}

/* Background blob — large organic accent shape */
.blob {
  position: absolute;
  width: 380px; height: 380px;
  border-radius: 60% 40% 30% 70% / 60% 30% 70% 40%;
  background: var(--accent);
  opacity: 0.18;
  pointer-events: none;
}
.blob-tl { top: -60px; left: -80px; }
.blob-br { bottom: -80px; right: -60px; transform: rotate(45deg); }

/* Stats area: rounded card on contrasting background */
.stats-section {
  background: rgba(255,255,255,0.07);
  border-radius: 28px;
  padding: 44px 48px;
  backdrop-filter: blur(4px);
}

/* Large decorative SVG illustration anchor */
.illus {
  position: absolute;
  opacity: 0.15;
  right: 48px; bottom: 48px;
  width: 180px; height: 180px;
  fill: var(--text);
}

SVG illustration guidance: build a simple but bold thematic silhouette in inline SVG. Examples:

• Wheat/grain trade: Three wheat stalks, side by side, with stylised grain heads — 5–8 <path> elements.
• Emissions/climate: A billowing cloud/smoke column rising from a horizon line.
• Food security/hunger: A stylised bowl with a spoon, or a globe with fork and knife flanking.
• Livestock: A side-profile cow or chicken silhouette, single solid fill.
• Temperature: A thermometer with a rising mercury column and radiating lines.

These do NOT need to be photorealistic — geometric, flat-style SVG in 2–3 path shapes is the right aesthetic. Fill with palette --hero or --text at 15–25% opacity for depth without distraction.

Screenshotting the card: use Playwright with viewport matching card dimensions exactly and deviceScaleFactor: 3 for retina-quality output (yields a 3240 × 4050 px PNG at 4:5):

from playwright.sync_api import sync_playwright
with sync_playwright() as p:
    browser = p.chromium.launch()
    page = browser.new_page(
        viewport={"width": 1080, "height": 1350},
        device_scale_factor=3,   # 3× retina — 3240×4050 output
    )
    page.goto(f"file://{path_to_social_html}")
    page.wait_for_load_state("networkidle")  # wait for Google Fonts

    # Force animation end-state before capture. For `full_page=False` social
    # cards this is mostly defensive (the card fits in viewport so the observer
    # usually fires), but it's required for `full_page=True` captures of tall
    # infographics where elements below the fold never enter the viewport.
    page.emulate_media(reduced_motion="reduce")
    page.evaluate(
        """() => document
          .querySelectorAll('.stat, .hero-stat, [data-draw], .trade-arc')
          .forEach(el => el.classList.add('visible'))"""
    )
    page.wait_for_timeout(100)

    page.screenshot(path=png_path, full_page=False)  # full_page=False = viewport only
    browser.close()

full_page=False is critical — it captures exactly the viewport, not the full scrollable document.

Export quality pipeline

Three tiers of output, in increasing quality order. Always use the highest tier available.

Tier 2 — Playwright page.pdf() (default for all PDF asks):

from playwright.sync_api import sync_playwright
with sync_playwright() as p:
    browser = p.chromium.launch()
    page = browser.new_page()
    page.goto(f"file://{path_to_html}")
    page.wait_for_load_state("networkidle")

    # CRITICAL for animated infographics: page.pdf() does not scroll, so
    # IntersectionObserver-driven animations never fire and animated elements
    # stay in their pre-animation hidden state (invisible chart fills, faded
    # stat cards, undrawn line-chart paths).
    # Emulating print media + reduced-motion triggers the JS early-return that
    # force-adds .visible to every animated element, and activates the
    # @media print CSS block that kills transitions and forces end-state.
    page.emulate_media(media="print", reduced_motion="reduce")
    # Defensive fallback in case the HTML was generated without the reduced-
    # motion early-return (older template or hand-written HTML).
    page.evaluate(
        """() => document
          .querySelectorAll('.stat, .hero-stat, [data-draw], .trade-arc')
          .forEach(el => el.classList.add('visible'))"""
    )
    page.wait_for_timeout(100)  # let the DOM repaint after class changes

    page.pdf(
        path=pdf_path,
        format="A4",              # or "Letter"
        print_background=True,    # CRITICAL — without this, all fills/gradients are stripped
        margin={"top": "0", "right": "0", "bottom": "0", "left": "0"},
    )
    browser.close()

Tier 3 — Inkscape CLI (upgrade silently when inkscape is installed; skip if not found):

import subprocess, shutil
from bs4 import BeautifulSoup

def export_via_inkscape(html_path, svg_path, pdf_path):
    # Extract the primary <svg> from the HTML and save standalone
    soup = BeautifulSoup(open(html_path).read(), "html.parser")
    svg = soup.find("svg")
    if not svg:
        return False  # no SVG to extract; fall back to Tier 2
    svg_path.write_text(str(svg))
    subprocess.run([
        "inkscape", str(svg_path),
        "--export-type=pdf",
        f"--export-filename={pdf_path}",
        "--export-dpi=300",
    ], check=True)
    return True

if shutil.which("inkscape"):
    ok = export_via_inkscape(html_path, svg_path, pdf_path)
if not shutil.which("inkscape") or not ok:
    # Fall back to Tier 2
    ...

Quality ceiling without external design tools:

With Tier 2/3, the output reaches 85–90% of premium data journalism studio quality. The remaining gap is hand-crafted illustration work: character figures, complex flow diagrams, and multi-element scene compositions. Those require a human designer or an interactive vector tool (Figma, Illustrator). Figma, Canva, and Adobe Express are not currently suitable for agent-driven layout generation — they require interactive editing sessions and have no practical programmatic "generate from data" API for arbitrary layouts.

Step 9 — Save and describe

Share a computer:// link to the HTML file and any opt-in exports. Give a 2–3 sentence description:

• The hero number and what it means.
• Which palette was chosen and why.
• Any invariant that materially shaped the output (China composite choice, rankings fallback).

Step 10 — Offer refinements

Offer four specific levers:

• Different hero stat ("what if we lead with the Africa–Europe intensity gap instead?").
• Different main visual ("swap the time series for a map?").
• Palette swap ("try Meadow instead of Ember?").
• Emphasis shift ("shrink the headline, grow the hero?").

Composition with other skills

• faostat-map — called when the main visual is a choropleth. Pass palette, disaggregate_china=true, output_format='svg'. Map skill returns an SVG string to paste inline.
• faostat-analytical-brief — separate deliverable. An infographic is not a slimmed-down brief; the brief keeps the dense information-per-pixel house style with a data appendix. Infographic is sparse, single-page, visual-first.
• faostat-story — sibling. A story is long-form prose with charts interspersed; an infographic collapses the same subject into one page. They can share data pulls but not layouts.

Error handling

• Empty faostat_get_data payload — retry with comma year list (invariant 2); if still empty, widen the year range by ±2 years and note the adjustment in the source footer.
• requires_confirmation on faostat_search_codes — stop and ask the user via AskUserQuestion. Do not guess.
• faostat_get_rankings HTTP 500 — fall back to faostat_get_data across all reporting countries and sort client-side. Note in the source footer: "Rankings reconstructed client-side".
• Playwright / Chromium unavailable — fall back to weasyprint for PDF; skip PNG and tell the user the sandbox lacks the headless-browser runtime.
• Google Fonts CDN blocked — inline a system font stack as fallback: -apple-system, 'Segoe UI', 'Helvetica Neue', Arial, sans-serif for body and Georgia, 'Times New Roman', serif for hero. Note the fallback.
• User asks to add FAO branding (logo, masthead, "Required citation: FAO") — push back. Explain that the infographic skill drops FAO branding by design so the output isn't mistaken for an FAO publication. The CC-BY-4.0 data attribution in the source footer is sufficient and legally required. Only proceed if the user explicitly overrides.
• User asks for multiple hero stats — push back. Invariant 8: one hero. Offer to turn the second would-be-hero into a supporting stat, or to build a second infographic.
• User asks for a second chart (rankings, second time series, composition + map, etc.) — push back. Invariant 9: one main visual. Offer (a) to promote the new chart to the main visual and demote the current one, (b) to render the ranking/composition as a plain numbered list or a small numeric table with no bar indicators, or (c) to split into two infographics.
• Urge to label a chart "AR5 GWP-100", "CO2eq", "kt", etc. — push back. Invariant 10: jargon belongs only in the source footer. Rewrite the label in plain English and add the methodology clause to the footer.

Suggested citation

Prepared by [Author]. [YYYY]. [Headline]. Infographic based on FAOSTAT data, accessed [Month YYYY]. Licence: CC-BY-4.0.

Not "Required citation: FAO. …". Any such line is a bug.

Related Skills