Graphic Chart Agent Skill

graphic-chart

Generates data visualization charts as PNG. Renders HTML with Apache ECharts v6 in headless Chromium via Playwright → screenshots at 2× retina quality.

CDN: https://cdn.jsdelivr.net/npm/[email protected]/dist/echarts.min.js

Critical Rules (read before every generation)

• area type → type: 'line' + areaStyle: {} — ECharts has no type: 'area'.
• doughnut → type: 'pie' + radius: ['40%', '70%'] — ECharts has no type: 'doughnut'.
• Readiness signal: register chart.on('finished', fn) BEFORE chart.setOption() — ECharts bug #14101/#17500: if listener is registered after setOption, it silently never fires. Always register both finished and rendered events before setOption.
• xAxis.type: 'category' must be explicit — ECharts does not infer it from the data. Forgetting this produces a blank chart.
• Category labels go in xAxis.data, not in a data.labels array. ECharts structure is flat: { xAxis, yAxis, series, grid, title, legend } — not nested under data or options.
• Data labels are fully built-in — use label: { show: true } on any series. No plugin needed.
• Highlight a specific bar/point via per-item itemStyle — put { value: N, itemStyle: { color: '#...' } } directly in the data array. Do NOT use Chart.js-style backgroundColor arrays.
• ECharts init uses a <div> container, not <canvas> — echarts.init(document.getElementById('chart')). The container div needs explicit dimensions.
• animation: false in option — disables animation for instant render. Still register finished + rendered events before setOption for the readiness signal.

10. Never dump HTML in chat. Save to file, show summary only.

11. Title states the insight, not the subject. "Revenue grew 3× in 12 months" not "Monthly Revenue".

12. Pie/doughnut: use body padding: 64px 80px and .chart-container { max-height: 860px } — prevents edge-to-edge fill when no title.

Step 1: Intake

Required: chart_type, data

Optional parameters and defaults:

| Parameter | Default | Description |

|---|---|---|

| chart_type | — | bar / line / area / pie / doughnut / scatter / radar / treemap |

| data | — | JSON array or CSV — required |

| title | — | States the insight, ≤10 words |

| subtitle | — | 1-sentence context line |

| style | clean-slate | clean-slate / midnight-editorial / matt-gray / electric-burst / brutalist |

| dimensions | 1080x1080 | WxH pixels (output PNG = 2× via deviceScaleFactor) |

| x_label | — | X-axis label text |

| y_label | — | Y-axis label text |

| source | — | Data source shown in footer |

| highlight | — | Data label to highlight (e.g. "Q4", "Dec", index 3) |

If chart_type or data is missing, ask exactly:

> "To create the chart, I need:

> 1. Chart type — bar / line / area / pie / doughnut / scatter / radar / treemap

> 2. Data — provide as JSON array or CSV (e.g. [12, 18, 22, 25, 31] with labels ['Q1','Q2','Q3','Q4','Q5'])

>

> Optional: title, style (default: clean-slate), dimensions (default: 1080×1080), highlight a specific data point"

If both present → skip to Step 2.

Step 2: Internal Architecture (never shown to user)

1. Normalize chart type:

• area → line + areaStyle: {} on series
• doughnut → pie + radius: ['40%', '70%'] on series
• horizontal bar → bar + swap xAxis/yAxis (category axis on y)
• All others: use as-is

2. Read references/chart-library.md — load full config spec for this chart type.

3. Read references/style-presets.md — load CSS tokens + data palette for chosen style.

4. Commit to design direction:

| Decision | Derive from |

|---|---|

| Tone | Professional / editorial / bold / technical — match the data's audience |

| Data story | Single insight this chart proves (becomes the title) |

| Highlight strategy | Which data point needs visual emphasis and why? |

| Background | Light (clean-slate, matt-gray) or dark (midnight-editorial, electric-burst, brutalist) |

5. Parse data:

• Simple array [12, 18, 22] → series.data, labels provided separately
• Object array [{x: 'Jan', y: 12}] → xAxis.data from x keys, series.data from y values
• CSV: parse header row as xAxis.data, value row as series.data
• Multi-series: multiple series entries each with type, name, data
• Scatter: series.data: [[x1,y1], [x2,y2], ...] format

6. Parse dimensions: "1080x1080" → W=1080, H=1080. Body = WxH. Output PNG = 2W × 2H.

Step 3: HTML Generation

Read ALL before generating:

• references/chart-library.md for this chart type's full ECharts config spec
• references/style-presets.md for the chosen style's CSS tokens + palette

Required HTML structure:

<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
[font CDN link from style preset]
<style>
:root {
  [all CSS tokens from style preset]
}

*, *::before, *::after { margin: 0; padding: 0; box-sizing: border-box; }
html, body {
  width: [W]px; height: [H]px;
  overflow: hidden;
  background: var(--bg);
  font-family: var(--font-body);
}
body {
  display: flex;
  flex-direction: column;
  padding: 40px 48px 32px;   /* pie/doughnut: use 64px 80px */
}

/* ECharts container must have explicit size */
.chart-container {
  flex: 1;
  min-height: 0;
  /* pie/doughnut only: max-height: 860px; */
}

.chart-header { margin-bottom: 24px; }
.chart-title {
  font-family: var(--font-display);
  font-size: clamp(1.1rem, 2.5vw, 1.6rem);
  font-weight: 700;
  color: var(--text);
  line-height: 1.2;
}
.chart-subtitle {
  font-family: var(--font-body);
  font-size: clamp(0.75rem, 1.2vw, 0.9rem);
  color: var(--text-muted);
  margin-top: 6px;
  line-height: 1.5;
}
.chart-footer {
  margin-top: 14px;
  font-family: var(--font-body);
  font-size: 10px;
  color: var(--text-muted);
  opacity: 0.65;
}
</style>
</head>
<body>

[if title or subtitle: <div class="chart-header"><div class="chart-title">...</div>...</div>]

<div id="chart" class="chart-container"></div>

[if source: <div class="chart-footer">Source: [source]</div>]

<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/echarts.min.js"></script>

<script>
window.__chartReady = false;

document.fonts.ready.then(() => {
  const container = document.getElementById('chart');
  const chart = echarts.init(container, null, { renderer: 'canvas' });

  // CRITICAL: register events BEFORE setOption — ECharts bug #14101/#17500
  // 'finished' may silently not fire if registered after setOption with animation:false
  chart.on('finished', () => {
    window.__chartReady = true;
  });
  // Belt-and-suspenders fallback via 'rendered'
  chart.on('rendered', () => {
    clearTimeout(window.__renderDebounce);
    window.__renderDebounce = setTimeout(() => { window.__chartReady = true; }, 100);
  });

  const palette = [palette from style preset];

  const option = {
    animation: false,                            // instant render for screenshot

    backgroundColor: 'transparent',             // body CSS handles bg color

    textStyle: {
      fontFamily: '[--font-body value]',
      color: '[--text-muted value]',
    },

    [title config if title param provided],
    [legend config per chart type],
    [grid config per chart type],
    [xAxis config per chart type],
    [yAxis config per chart type],

    series: [{
      [full series config from chart-library.md for this type]
      [palette colors applied per chart type]
      [if highlight: per-item itemStyle on the highlighted data point]
    }]
  };

  chart.setOption(option);   // setOption ALWAYS comes after event registration
});
</script>
</body>
</html>

Design quality rules:

• Title font: fontWeight: 'bold', fontSize: 20–24 — no thin titles
• Grid lines: low opacity (0.06–0.10) — subordinate to data
• Bars: barMaxWidth: 60, rounded via itemStyle.borderRadius: [4,4,0,0]
• Lines: smooth: true for natural curves, symbolSize: 8 for points
• Pie/doughnut: label.formatter: '{b}\n{d}%' for built-in on-slice labels
• Dark presets: grid rgba(255,255,255,0.07), axis line/tick color rgba(255,255,255,0.15)
• Tooltip: show: false — static PNG, no hover interaction
• When no title provided: skip .chart-header, omit title from ECharts option

Step 4: Self-QA (fix every failure before Step 5)

Structure:

• [ ] Container is a <div>, not <canvas>
• [ ] window.__chartReady = false declared before document.fonts.ready
• [ ] chart.on('finished', ...) registered BEFORE chart.setOption()
• [ ] chart.on('rendered', ...) debounce fallback registered BEFORE chart.setOption()
• [ ] animation: false in option object
• [ ] chart.setOption(option) is the LAST call in the init block

Type-specific:

• [ ] Area: type: 'line' + areaStyle: {} — no type: 'area'
• [ ] Doughnut: type: 'pie' + radius: ['40%', '70%'] — no type: 'doughnut'
• [ ] Bar: xAxis.type: 'category' explicitly set
• [ ] Category data in xAxis.data (not data.labels)
• [ ] Pie/doughnut: body padding: 64px 80px + .chart-container { max-height: 860px }
• [ ] Highlight: per-item { value: N, itemStyle: { color } } in data array

Design:

• [ ] All palette colors from references/style-presets.md
• [ ] Title states insight (not just subject)
• [ ] tooltip: { show: false } or omitted (no hover on static PNG)
• [ ] Dark preset: grid/axis colors use rgba(255,255,255,...)
• [ ] Source in footer if source param provided
• [ ] Data labels visible (built-in label: { show: true } on series)

Step 5: Export

Determine slug from title or chart type + data context (kebab-case, ≤30 chars):

mkdir -p chart/[slug]

Save HTML: chart/[slug]/chart.html

Quick browser check:

open chart/[slug]/chart.html

Run export (replace [skill-root] with actual path to this skill's directory):

bash [skill-root]/scripts/export-chart.sh \
  chart/[slug]/chart.html \
  chart/[slug]/chart.png \
  --width [W] \
  --height [H]

The script installs Playwright on first run (~200MB Chromium download), then captures the chart at deviceScaleFactor: 2.

Step 6: Output Summary

## Chart: [title]
Date: [YYYY-MM-DD] | Type: [chart_type] | Style: [style]
Dimensions: [W×H]px → PNG: [2W×2H]px @2× retina

Files
  Source:   chart/[slug]/chart.html
  Output:   chart/[slug]/chart.png
  Size:     [X] KB

Checklist
- [ ] Title states the insight clearly
- [ ] Data labels legible at display size
- [ ] Highlight visible on correct data point
- [ ] Source attribution present in footer

Prompt Tips (show when user asks for guidance)

> "Provide structured data — JSON or CSV, not prose descriptions."

>

> "Name the chart type explicitly. 'bar chart comparing Q1–Q4' not 'a chart showing quarters'."

>

> "Specify the data story. 'highlight Q4 which outperformed all others' gives the annotation context."

>

> ✅ Good: "Create a line chart. Title: 'From $12k to $95k ARR in 12 Months'. Data: [12, 18, 22, 25, 31, 38, 44, 52, 61, 68, 78, 95] (Jan–Dec 2024). Highlight December. Source: Internal CRM. Style: electric-burst."

>

> ❌ Bad: "make a chart about our company growth"