microsoft-agent-framework

Microsoft Agent Framework

Use this skill when working with applications, agents, workflows, or migrations built on Microsoft Agent Framework.

Microsoft Agent Framework is the unified successor to Semantic Kernel and AutoGen, combining their strengths with new capabilities. Because it is still in public preview and changes quickly, always ground implementation advice in the latest official documentation and samples rather than relying on stale knowledge.

Determine the target language first

Choose the language workflow before making recommendations or code changes:

1. Use the .NET workflow when the repository contains .cs, .csproj, .sln, .slnx, or other .NET project files, or when the user explicitly asks for C# or .NET guidance. Follow references/dotnet.md.
2. Use the Python workflow when the repository contains .py, pyproject.toml, requirements.txt, or the user explicitly asks for Python guidance. There is no Python reference file yet — ground Python guidance in the live official docs (see below) and the shared guidance in this skill.
3. If the repository contains both ecosystems, match the language used by the files being edited or the user's stated target.
4. If the language is ambiguous, inspect the current workspace first and then choose the closest language-specific reference.

Always consult live documentation

• Read the Microsoft Agent Framework overview first: https://learn.microsoft.com/agent-framework/overview/agent-framework-overview
• Prefer official docs and samples for the current API surface.
• Use the Microsoft Docs MCP tooling when available to fetch up-to-date framework guidance and examples.
• Treat older Semantic Kernel or AutoGen patterns as migration inputs, not as the default implementation model.

Shared guidance

When working with Microsoft Agent Framework in any language:

• Use async patterns for agent and workflow operations.
• Implement explicit error handling and logging.
• Prefer strong typing, clear interfaces, and maintainable composition patterns.
• Use DefaultAzureCredential when Azure authentication is appropriate.
• Use agents for autonomous decision-making, ad hoc planning, conversation flows, tool usage, and MCP server interactions.
• Use workflows for multi-step orchestration, predefined execution graphs, long-running tasks, and human-in-the-loop scenarios.
• Support model providers such as Azure AI Foundry, Azure OpenAI, OpenAI, and others, but prefer Azure AI Foundry services for new projects when that matches user needs.
• Use thread-based or equivalent state handling, context providers, middleware, checkpointing, routing, and orchestration patterns when they fit the problem.