data-craft

Beautiful, themed HTML data displays and briefings for coding agents.

data-craft is a skill for turning messy data, CSVs, JSON, notes, links, and topics into self-contained HTML reports, dashboards, tearsheets, visual briefings, and chart essays.

Most chart tools make charts.

data-craft makes publishable data briefings.

Each theme has its own visual language and writing voice: a Trading Floor tearsheet should not just look like a Bloomberg terminal; it should read like one. A Sand Hill essay should not just use a cream background; it should feel like a longform market brief. A Helvetica Riot broadsheet should have attitude, not generic AI pastel fog.

No build step. No backend. One HTML file. Open it, share it, host it, print it, or export it.

Gallery first

Pick a visual direction before you generate the piece.

What makes it different

data-craft is not a slide generator and not a generic dashboard template pack.

It is an AI-native analyst publishing system:

• Theme + voice are paired. The same topic reads differently in Sand Hill, Trading Floor, or Pitch Black.
• Data integrity is part of the format. Estimates must be labeled, stale figures must be verified, and every piece ends with methodology and sources.
• The output is portable. Each briefing is a single self-contained HTML file using ECharts from a CDN.
• The examples are real pieces. The repo ships six complete examples, not lorem ipsum placeholders.

If slides are presentations rebuilt for coding agents, data-craft is data display and briefing rebuilt for coding agents.

Quick start

Clone the repo:

git clone https://github.com/yixiliu617/data-craft.git

Then install or reference it in your agent.

Claude Code

git clone https://github.com/yixiliu617/data-craft.git ~/.claude/skills/data-craft

Prompt:

Use the data-craft skill. Build a Sand Hill chart essay on AI coding agents.

Claude.ai

Upload the data-craft/ folder to a Project and add this to the project instructions:

Use the data-craft skill whenever I ask for a chart, dashboard, data display, briefing, tearsheet, or report.

Codex / other coding agents

Point the agent at this repo and the skill file:

Read ./SKILL.md and use data-craft to generate a self-contained HTML briefing.
Use the relevant theme card in ./themes before writing the output.

For Codex-style local workflows, keep this repo in your workspace and reference it directly:

Use C:\path\to\data-craft\SKILL.md.
Create a Trading Floor tearsheet from this CSV.

More install notes: docs/agent-install.md.

Example prompts

Let data-craft pick the theme

Use data-craft to build a briefing on the AI coding tools market.
Include market structure, product comparison, GitHub stars, and a methodology footer.

Specify the theme

Build a Trading Floor tearsheet on Nvidia's latest earnings.
Use current data, label estimates, and include a dense KPI table.

Convert a CSV into a briefing

Use data-craft with the Sand Hill theme.
Turn this CSV into a 900-word chart essay with two figures, a source footer, and three takeaways.

[paste CSV]

Turn a topic into a visual report

Build an Apollo White briefing on AI data centers.
Compare GPUs, networking, power, cooling, and capex bottlenecks.
Use verified current sources and label estimates.

More examples: docs/examples.md.

Visual preview selection

When you are unsure which theme fits, ask the agent to show a preview menu first:

Before generating the full piece, show me the six data-craft theme options and recommend one.
Explain the tradeoff in one sentence per theme.

Recommended workflow:

1. Pick a theme.
2. Generate a 1-screen skeleton.
3. Confirm the direction.
4. Generate the full HTML.
5. Render or screenshot-check before publishing.

Full guide: docs/visual-preview-selection.md.

Open the static preview gallery:

examples/theme-preview.html

Export to PDF

Every data-craft HTML file can be printed from the browser.

For automated export, use Playwright or another headless browser to render the HTML and print to PDF. See:

docs/export-pdf.md

Optional helper script:

node scripts/export-pdf.mjs examples/ai-coding-tools.html ai-coding-tools.pdf

The script expects Playwright to be installed in your local environment.

Demo video

Watch/download the silent MP4: assets/data-craft-demo-silent.mp4

A short demo should show:

1. Pick a topic.
2. Choose a theme.
3. Generate the HTML.
4. Open the output.
5. Show charts, methodology, and export/share options.

Script and shot list: docs/demo-video-script.md.

Browser-based recorder:

examples/demo-video-recorder.html

Project structure

data-craft/
  README.md
  SKILL.md
  themes/
    sand-hill.md
    pitch-black.md
    trading-floor.md
    helvetica-riot.md
    sunday-press.md
    apollo-white.md
  examples/
    theme-preview.html
    demo-video-recorder.html
    ai-coding-tools.html
    humanoid-robotics.html
    nvda-tearsheet.html
    ai-music-wars.html
    d2c-coffee.html
    spacex-ipo.html
  assets/
    gallery-hero.png
    thumb-*.png
    data-craft-demo.gif
    data-craft-demo-silent.mp4
    demo-source/
  docs/
    agent-install.md
    examples.md
    export-pdf.md
    visual-preview-selection.md
    demo-video-script.md
    methodology.md
    prompting-guide.md
  scripts/
    export-pdf.mjs
    capture-theme-fullpages.mjs
    create_demo_gif.py
    create_demo_mp4.py
    export-demo-pages.py

Roadmap

Now

• Six themes.
• Six golden examples.
• Single-file HTML output.
• Theme and voice rules.
• Methodology and estimate-labeling conventions.

Next

• Visual preview selection workflow.
• Export-to-PDF helper.
• More CSV and topic-to-briefing examples.
• Short demo video.
• Better install paths for Claude, Codex, and other agents.

Later

• Optional slide mode for briefing-to-deck conversion.
• More finance-native themes.
• Custom brand/theme adapters.
• Component library for embedding charts and tiles into other sites.

Slide mode should stay downstream of the core mission: data-craft is first a research briefing system, not a presentation tool.

License

MIT. See LICENSE.

data-craft is independent open-source software. It is not affiliated with Anthropic, Bloomberg, SpaceX, or any publication, company, or tool referenced as design inspiration.