flint-theme-author

作者: microsoft

使用時機:從品牌指南、網站、簡報、出版物…建立、翻譯、精煉、審查或驗證自訂 Flint ThemeSpec

npx skills add https://github.com/microsoft/flint-chart --skill flint-theme-author

Flint ThemeSpec authoring

Create a reusable visual system for Flint charts. Translate the user's reference materials into one valid ThemeSpec that can be pasted directly into Theme Lab.

Authoritative references

Read these before authoring. When prose and source disagree, the TypeScript contract is authoritative.

Do not invent keys. If a requested decision cannot be represented by ThemeSpec, omit it and briefly identify the unsupported decision outside the JSON only when the user asked for an explanation.

Output contract

Unless the user asks for commentary, return exactly one valid JSON object:

  • Return the bare ThemeSpec, not { "theme_spec": ... } and not a complete ChartAssemblyInput.
  • Do not wrap JSON in a Markdown fence.
  • Do not include comments, placeholders, ellipses, or trailing commas.
  • Include a stable kebab-case id and a human-readable label.
  • Include only supported fields and only decisions justified by the references.
  • Omit unspecified optional fields rather than guessing.

The top-level output schema is:

{
  "extends": "optional-preset-id",
  "id": "theme-id",
  "label": "Theme label",
  "ink": {},
  "type": {},
  "structure": {},
  "marks": {},
  "labels": {},
  "legend": {},
  "dataLabels": {},
  "annotation": {},
  "furniture": [],
  "facets": {},
  "layout": {},
  "chartDefaults": {},
  "compileDefaults": {},
  "interaction": {},
  "variants": []
}

Every block except id and label may be omitted. The skeleton shows the shape, not a requirement to emit empty blocks.

Theme boundary

A theme controls presentation and compiler behavior across many charts. It may decide color, typography, surfaces, grids, axes, mark geometry, labels, legends, annotations, density, spacing, facets, and reusable chart defaults.

A theme does not choose data fields, encodings, aggregation, filtering, sorting, chart titles, or data values. Never put those decisions in a ThemeSpec. Outside chartDefaults, authored policy must not name a chart type, positional channel, mark type, field, or backend property.

Workflow

  1. Inspect the supplied references. Extract recurring decisions, not isolated decoration from one screenshot.
  2. Ask only for missing decisions that materially affect the theme: identity, required colors, typography, surface/background, density, accessibility constraints, and whether an existing Flint preset is a useful base.
  3. Separate evidence into system roles:
    • surfaces and text hierarchy;
    • categorical, sequential, diverging, and status color;
    • typography roles;
    • axes, grids, baselines, frames, and structural ink;
    • mark weight, corners, points, separators, and spacing;
    • labels, legends, annotations, facets, and layout behavior.
  4. Choose standalone authoring or inheritance.
  5. Author the smallest coherent spec that expresses the system.
  6. Check every key and enum against types.ts.
  7. Return the bare JSON object.

Choosing inheritance

Use extends when a built-in preset already supplies the intended compiler behavior and the new theme is a focused variation. Available preset IDs are:

nyt, economist, swiss, nature, mckinsey, datawrapper, powerbi, powerbi-light, pop, and cartoon.

Nested objects merge. Arrays and scalar values replace the preset value. If you replace ink.series.categorical, also consider replacing categoricalExtended; otherwise high-cardinality charts may return to the base preset's extended palette.

Use a standalone theme when the reference system does not honestly inherit a preset's layout, typography, and presentation behavior.

Schema map

Use the source contract for exhaustive nested fields and enum values. These are the authored blocks and their jobs:

BlockPurpose
inkCanvas, plot, panel, text, structural, series, status, and accent colors
typeMinimum size and role-based typography for headlines, axes, values, annotations, footnotes, and displays
structureSemantic categorical/measure axes, grids, baseline, and frame
marksBand width, strokes, interpolation, opacity, corners, outlines, points, connectors, intervals, summaries, and separators
labelsTruncation, flush behavior, and angle policy
legendVisibility, ordered placement choices, direction, title, swatches, and redundancy suppression
dataLabelsVisibility, placement, and ink behavior for values on marks
annotationUnits, axis titles, number formatting, point emphasis, labels, and statistics
furnitureRepeating masthead tabs and header/footer rules
facetsHeaders, frames, axis repetition, spacing, columns, and scale sharing
layoutDensity, target width, title block, and band step
chartDefaultsOptional defaults keyed by registered chart type or *; caller values still win
compileDefaultsPreferred base size, canvas size, and supported assemble options
interactionTooltip format
variantsConditional policy adaptations; variants may not change ink or type

High-value nested shapes

{
  "ink": {
    "surface": { "source": "house", "canvas": "#ffffff", "plot": "#ffffff", "panel": "#f5f5f5" },
    "text": { "primary": "#111111", "secondary": "#444444", "muted": "#777777", "inverse": "#ffffff" },
    "structure": { "axis": "#222222", "grid": "#dddddd", "frame": "#cccccc", "rule": "#222222", "zero": "#222222", "connector": "#888888" },
    "series": {
      "single": "#0067b8",
      "categorical": ["#0067b8", "#d83b01", "#107c10"],
      "categoricalExtended": ["#0067b8", "#d83b01", "#107c10", "#5c2d91"],
      "overflow": "#777777",
      "sequential": { "stops": ["#e8f3fb", "#0067b8"], "space": "lab", "endpointsAgainstSurface": true },
      "diverging": { "stops": ["#0067b8", "#f5f5f5", "#d83b01"], "neutral": "#f5f5f5", "space": "lab", "endpointsAgainstSurface": true },
      "status": { "positive": "#107c10", "negative": "#d83b01", "neutral": "#777777" }
    },
    "accent": "#0067b8"
  }
}

This is a shape example, not a palette recommendation. Derive actual values from the user's references.

Typography roles accept family, size, weight, style, case, and color. Allowed weights are regular, medium, semibold, and bold. Use available font families or defensible fallback stacks; do not claim a font is available merely because it appears in an image.

Presence-driven policies use omit, hairline, quiet, full, or emphasised. Density uses compact, normal, or airy. Read types.ts before using less common enums.

Translation guidance

  • Map brand neutrals to surfaces, text hierarchy, and structure before mapping accent colors to data series.
  • Treat categorical colors as an ordered identity set. Keep adjacent colors distinguishable and ensure the palette remains legible on the plot surface.
  • Build sequential ramps with ordered lightness. Build diverging ramps around a meaningful neutral. Do not use categorical arrays as numeric ramps.
  • Use status colors only for semantic positive, negative, and neutral meaning.
  • Preserve readable contrast for primary text, axes, labels, and values.
  • Translate recurring geometry into marks; do not encode a one-off shape from a single chart as global policy.
  • Prefer semantic structure (measure versus category) over physical x/y assumptions.
  • Keep the system coherent across line, bar, area, pie, matrix, distribution, multiseries, and faceted charts.
  • Use variants only for real conditional policy and include because to state why the adaptation exists.

Validation checklist

Before returning the JSON, verify:

  • It parses as strict JSON.
  • It is a bare object with id and label.
  • Every top-level and nested key exists in types.ts.
  • Every enum value and value type matches the source contract.
  • It contains no chart data, fields, encodings, titles, or backend JSON.
  • Text and structural ink remain readable against their surfaces.
  • Categorical colors are distinguishable; ramps have ordered lightness and usable endpoints.
  • Inherited arrays are intentionally replaced.
  • The theme expresses reusable rules rather than one chart's decoration.
  • The result can be pasted directly into Flint Theme Lab.

來自 microsoft 的更多技能

oss-growth
microsoft
開源增長駭客角色
agent-framework-azure-ai-py
microsoft
使用Microsoft Agent Framework Python SDK(agent-framework-azure-ai)构建Azure AI Foundry代理。适用于使用AzureAIAgentsProvider创建持久化代理、使用托管工具(代码解释器、文件搜索、网络搜索)、集成MCP服务器、管理对话线程或实现流式响应。涵盖函数工具、结构化输出和多工具代理。
development
airunway-aks-setup
microsoft
在AKS上設定AI Runway——從裸叢集到執行模型。涵蓋叢集驗證、控制器安裝、GPU評估、供應商設定及首次部署。時機:「設定AI Runway」、「上線AKS叢集」、「安裝AI Runway」、「airunway設定」、「部署模型至AKS」、「在AKS上進行GPU推論」、「在AKS上設定KAITO」、「在AKS上執行LLM」、「在AKS上使用vLLM」、「在AKS上設定模型服務」、「AI Runway控制器」。
devops
appinsights-instrumentation
microsoft
使用Azure Application Insights檢測Web應用程式的指南。提供遙測模式、SDK設定與組態參考。適用時機:如何檢測應用程式、App Insights SDK、遙測模式、什麼是App Insights、Application Insights指南、檢測範例、APM最佳實踐。
devops
applicationinsights-web-ts
microsoft
使用Application Insights JavaScript SDK(@microsoft/applicationinsights-web)為瀏覽器/Web應用程式進行檢測。適用於真實使用者監控(RUM)——頁面檢視、點擊、AJAX/fetch依賴、例外、自訂事件,以及與後端OpenTelemetry追蹤關聯的瀏覽器端GenAI代理追蹤。涵蓋SDK載入器指令碼與npm設定、框架擴充(React、React Native、Angular)、點擊分析、遙測初始化器,以及從瀏覽器發出的代理/工具/模型span的OTel GenAI語意慣例。
devops
azure-ai-anomalydetector-java
microsoft
使用適用於 Java 的 Azure AI 異常偵測器 SDK 建置異常偵測應用程式。在實作單變量/多變量異常偵測、時間序列分析或 AI 驅動監控時使用。
development
azure-ai-language-conversations-py
microsoft
使用 azure-ai-language-conversations Python SDK 實作對話語言理解(CLU)。當使用 ConversationAnalysisClient 分析對話意圖與實體、建置 NLP 功能,或將語言理解整合至應用程式時使用。
development
azure-ai-ml-py
microsoft
Azure Machine Learning SDK v2 for Python。用於機器學習工作區、作業、模型、資料集、計算資源與管線。 觸發詞:「azure-ai-ml」、「MLClient」、「workspace」、「model registry」、「training jobs」、「datasets」。
development