architecture-diagrams

作者: microsoft

雲端基礎設施的架構圖製作:解析 Azure IaC,對應關聯關係,並根據情況生成 ASCII 區塊圖或 Mermaid 流程圖…

npx skills add https://github.com/microsoft/hve-core --skill architecture-diagrams

Architecture Diagrams Skill

Goal

Turn infrastructure source files or a declared DS_CATALOG_V1 data model into a readable architecture diagram for reviews, ADRs, and design discussions. Preserve the caller's selected output format and the source's authority boundaries.

Infrastructure inputs include Terraform, Bicep, ARM templates, shell scripts, Kubernetes manifests, and Docker or Compose files. Catalog input uses declared entities and relationships from data-catalog, the durable data-catalog skill. It does not infer a data model from SQL or ORM files.

This skill documents infrastructure topology and data models. To document a software system, meaning its containers, its components, and the people and systems around it, use the c4-architecture skill instead.

Success criteria

  • The diagram includes only the confirmed source scope.
  • Infrastructure sources retain their existing parsing and relationship behavior.
  • Catalog diagrams preserve declared entity IDs, endpoints, cardinality, endpoint minimums, join keys, confidence, and evidence basis without inventing relationships.
  • Caller preference controls ASCII or Mermaid output.
  • Inferred and assumed catalog relationships remain visibly distinct from confirmed relationships.

Constraints

  • Treat a diagram as a view over source authority, not a semantic authority of its own.
  • Read catalog-erd.md for DS_CATALOG_V1 input, multiplicity mapping, confidence rendering, the catalog output contract, and the Functional Planner compatibility boundary.
  • Do not parse SQL DDL, Prisma, SQLAlchemy, or another ORM as catalog input.
  • Do not render primary-key, foreign-key, or uniqueness markers for catalog join keys; the catalog declares field names, not database key roles.
  • Keep the Feasibility Study Interchange Profile and downstream requirement mappings in their owning workstreams.

Stop rules

  • Stop and ask for scope when infrastructure boundaries are ambiguous.
  • Stop and report an unsupported catalog version, unresolved endpoint, unknown cardinality, missing or invalid endpoint minimum, unknown confidence value, or malformed join-key declaration instead of guessing or partially rendering.
  • Stop before diagram generation when no output preference can be resolved.

Output Format

This skill produces either ASCII block diagrams or Mermaid diagrams. Neither is the default: the caller or surrounding context chooses the output format for each diagram. When the caller does not state a preference, ask which format they want before generating.

The diagram type follows the source type. Infrastructure sources render as ASCII block diagrams or Mermaid flowcharts using the ASCII Conventions or Mermaid Conventions below. Catalog sources render as ASCII entity lines or a Mermaid erDiagram using catalog-erd.md. In every case, keep the structure, boundaries, and relationships identical across formats.

Preference Contract

Architecture diagram format selection applies beyond ADRs. For standalone usage, consult the root state file at .copilot-tracking/architecture-diagrams/state.json. If that file is absent, create it when a format is chosen.

{
  "userPreferences": {
    "diagramFormat": "mermaid"
  },
  "repoVisibility": "private"
}

The userPreferences.diagramFormat value must be either ascii or mermaid. The repoVisibility field is optional and may be used by surrounding workflows when they need to distinguish public and private repositories. Resolution order is:

  1. Explicit request or caller state.
  2. Root state file at .copilot-tracking/architecture-diagrams/state.json.
  3. If no preference is known for a standalone request, ask once and persist the answer to the root state file for later reuse.

Workflow

Follow this sequence when authoring a diagram:

  1. Discovery. Identify the relevant infrastructure files or declared catalog and the architectural scope. When the scope is unclear, ask which folders, services, or entities should be included.
  2. Parsing. For infrastructure, extract services, data stores, networking components, ingress points, and deployment units. For a catalog, execute scripts/render_catalog_erd.py or follow catalog-erd.md without adding inferred semantics.
  3. Relationship mapping. For infrastructure, connect components with the correct direction. For a catalog, retain the exact declared endpoints, cardinality, endpoint minimums, join keys, confidence, and basis.
  4. Generation. Render the final diagram in the caller's chosen format, with clear boundaries or entity labels and a compact legend.

ASCII Conventions

Use consistent box notation and alignment:

+------------------+      +------------------+
|   Service Name   |----->|   Service Name   |
+------------------+      +------------------+

Use the following conventions for readability:

  • Keep box labels short and specific.
  • Keep arrows aligned and use one relationship per line when possible.
  • Prefer clearly named boundaries over dense decoration.
  • Use repeated box shapes for similar components.

Arrow Types

ArrowMeaning
---->Data flow or dependency
<--->Bidirectional connection
- - >Optional or conditional resource

Grouping and Boundaries

Group related components inside a larger boundary when they share a network, account, or deployment domain.

Use a full box for a strong boundary:

+-----------------------------------------------+
|  Resource Group                               |
|                                               |
|  +-------------+        +-------------+       |
|  |   VNet      |------->|   Subnet    |       |
|  +-------------+        +-------------+       |
|                                               |
+-----------------------------------------------+

Use labeled boundaries for secondary or nested boundaries:

:--- Virtual Network ---------------------------:
:                                               :
:  +-------------+        +-------------+       :
:  |   Subnet A  |------->|   Subnet B  |       :
:  +-------------+        +-------------+       :
:                                               :
:-----------------------------------------------:

Mermaid Conventions

When the caller chooses Mermaid output, render a mermaid fenced code block using a flowchart that expresses the same structure, boundaries, and relationships you would draw in ASCII.

  • Use flowchart TB for top-to-bottom topologies and flowchart LR when the main flow reads left to right.
  • Declare each component as a node with a short, specific label, for example lb["Load Balancer"], and use [("...")] for data stores.
  • Group components that share a network, account, or deployment domain inside a subgraph block, such as a VNet, subnet, or resource group.
  • Use --> for data flow or dependency, <--> for bidirectional connections, and -. optional .-> for optional or conditional links.
  • Keep node identifiers stable and lowercase, and reserve labels for the human-readable name.
flowchart TB
    subgraph rg["Resource Group"]
        lb["Load Balancer"]
        subgraph subnet["App Subnet"]
            vm1["VM 1"]
            vm2["VM 2"]
        end
        db[("SQL Database")]
    end
    lb --> vm1
    lb --> vm2
    vm1 --> db
    vm2 --> db

Layout Guidelines

  • Place external or public services at the top.
  • Place compute or application tiers in the middle.
  • Place data stores at the bottom.
  • Group components by network boundary, such as a VNet or subnet.
  • Let the main flow run from top to bottom when the direction is clear.

Resource Identification Heuristics

When reading infrastructure sources, extract:

  • Resource type and name
  • Network associations, including VNet, subnet, private endpoint, or ingress settings
  • Dependencies that are explicit in configuration and those that are implied by references
  • Deployment relationships such as container registry, service mesh, or workload placement

Output Format Contract

Use this structure for every infrastructure diagram:

## <Name> Architecture

[diagram in the selected format]

### Legend
[Arrow meanings from this diagram; reference the arrow types above]

### Key Relationships
[Notable connections and dependencies]

The title should use title case and follow the pattern <Name> Architecture. The legend should explain any special symbols used, and the key relationships section should focus on the most important dependencies or data flows.

Catalog diagrams use the parallel ## <Engagement> Data Model contract defined in catalog-erd.md, with a Legend covering multiplicity and confidence and a Key Relationships section carrying every declared relationship and its basis.

Worked Example: AKS Platform Architecture

## AKS Platform Architecture

+===============================================================+
|  Resource Group                                               |
|  :--- Virtual Network ------------------------------------:   |
|  :  +------------------+        +------------------+      :   |
|  :  |   NAT Gateway    |------->|   AKS Cluster    |      :   |
|  :  +------------------+        +--------+---------+      :   |
|  :                              +--------v---------+      :   |
|  :                              |       ACR        |      :   |
| :                              +------------------+      : |
|:----------------------------------------------------------:|
|      +------------------+        +------------------+      |
|     | Log Analytics    |<-------|  App Insights    |          |
|     +------------------+        +------------------+          |
+===============================================================+

### Legend
See the arrow types above. Additional symbols: `====` primary boundary, `:---:` secondary boundary.

### Key Relationships
* AKS pulls images from ACR through the network boundary.
* NAT Gateway provides egress for AKS workloads.

The same architecture in Mermaid form expresses identical structure, boundaries, and relationships:

## AKS Platform Architecture

```mermaid
flowchart TB
    subgraph rg["Resource Group"]
        subgraph vnet["Virtual Network"]
            nat["NAT Gateway"]
            aks["AKS Cluster"]
            acr["ACR"]
        end
        appinsights["App Insights"]
        logs[("Log Analytics")]
    end
    nat --> aks
    aks --> acr
    appinsights --> logs
```

### Legend
See the arrow types above; `subgraph` blocks denote network or resource boundaries.

### Key Relationships
* AKS pulls images from ACR through the network boundary.
* NAT Gateway provides egress for AKS workloads.

Authoring Guidelines

  • Ask one or two clarifying questions when the architecture scope is ambiguous.
  • Announce the current workflow stage when you move from discovery to parsing or generation.
  • Present a draft diagram with a short summary of the resources included before finalizing.
  • Note important inference decisions, such as implicit dependencies, when they affect the diagram.
  • Treat the diagram as a static representation of infrastructure sources, not a runtime execution view.
  • Keep the output focused on a single architecture scope so it remains readable.

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