菜单
Conceptual organization of Agent Framework documentation and samples. For language-specific details (file naming, provider setup, code patterns), see each code repo's skills files.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/MicrosoftDocs/semantic-kernel-docs --skill sample-structure命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | sample-structure |
| description | Conceptual organization of Agent Framework documentation and samples. For language-specific details (file naming, provider setup, code patterns), see each code repo's skills files. |
This skill defines the conceptual organization of Agent Framework documentation and how it maps to the learning journey. Language-specific implementation details belong in each code repo's skills files — this file covers the shared conceptual model only.
The Agent Framework uses a 5-layer structure that progressively builds complexity. Both docs and code samples mirror this structure so users can move fluidly between reading and running code.
Layer 1: Get Started → Linear tutorial, one concept per step
Layer 2: Agent Concepts → Deep-dive reference, organized by topic
Layer 3: Workflows → Multi-step orchestration patterns
Layer 4: Hosting → Deployment, protocols, production infrastructure
Layer 5: End-to-End → Complete applications combining all layers
| Layer | Purpose | Complexity | Audience |
|---|---|---|---|
| Get Started | "Follow these steps" — sequential, cumulative | Beginner | New users |
| Agent Concepts | "Go deeper on X" — one topic per page | Intermediate | Users building features |
| Workflows | "Orchestrate multiple steps" — pattern-based | Intermediate | Users composing agents |
| Hosting | "Deploy to production" — infrastructure-focused | Advanced | Users shipping |
| End-to-End | "See it all together" — reference applications |
| Advanced |
| Users architecting |
Each step adds exactly one concept to the previous step:
Concepts are grouped by architectural concern:
The Agent Framework uses a layered middleware architecture with three interception points:
Agent.run() call; can modify messages, options, threadEach layer has its own context object and supports call_next() for pipeline chaining.
Middleware can return normally (upstream post-processing runs), raise MiddlewareTermination
(skips all post-processing), or raise exceptions (propagated to caller).
Every docs page under get-started/ maps 1:1 to a sample file in both Python and .NET.
Deep-dive docs pages link to corresponding concept samples.
| Docs Section | Code Repo Layer | Cross-link pattern |
|---|---|---|
get-started/*.md | 01-get-started/ | Each step → next step + lateral deep-dive |
agents/**/*.md | 02-agents/ | Each topic → matching concept sample(s) |
workflows/*.md | 03-workflows/ | Each pattern → matching workflow sample |
integrations/**/*.md | Current location depends on the feature | Keep existing sample references until the corresponding phased sample move lands |
hosting/**/*.md | 04-hosting/ | Hosting guides → matching hosting sample |
The documentation taxonomy can move before the sample taxonomy. During that
transition, move docs and update doc-to-doc links, TOC entries, and redirects,
but keep :::code paths and sample repository URLs unchanged until the sample
move is available.
Docs pages follow consistent navigation:
agent-framework/AGENTS.md, then
determine which learning layer the content belongs to