microsoft/agent-framework
7 skills297 installs78.1k starsGitHub
Install
npx skills add https://github.com/microsoft/agent-frameworkSkills in this repo
1Python Code Qualitypython-code-quality is a procedural agent skill for Microsoft’s Agent Framework Python workspace. It catalogs the exact uv/poe commands solo builders and small teams need to format code, run Ruff or syntax pipelines, fan out Pyright and MyPy across packages, execute parallel package checks, and mirror full CI locally—including samples and markdown code blocks. Use it whenever you are greening a branch, reproducing a failed GitHub Action, or installing prek hooks so commits stay lightweight while still catching regressions. The skill assumes you work from the python/ root and names flags for targeting core packages, all packages (-A), or samples-only paths. It does not replace product feature work; it standardizes how an coding agent executes checks without guessing script names or directory layout, which matters for agent repos where one wrong path wastes an entire agent turn.90installs2Python DevelopmentPython Development is an agent skill that encodes how Microsoft expects Python to look inside the Agent Framework repository. Solo builders and contributors invoking it while modifying files under python/ get consistent copyright headers, strict annotation rules, and parameter design patterns that keep public APIs readable for both humans and static analyzers. It favors union syntax over Optional, Mapping for read-only inputs, and explicit string overrides where enums would force extra imports—illustrated with create_agent-style tool_mode handling. The guidance also covers when to use targeted type ignores versus casts, and how to structure functions so mypy and pyright stay aligned. Use it whenever an coding agent drafts or refactors framework Python so reviews focus on behavior instead of style drift. It does not teach agent architecture from zero and is scoped to this repo’s standards rather than every Python project.51installs3Build And TestBuild-and-test is a procedural agent skill for contributors working in Microsoft’s Agent Framework .NET repository. Solo or small-team developers use it when an agent needs to verify a patch: restore dependencies, build a specific Microsoft.Agents.AI package, run unit tests with precise filters, or apply dotnet format without guessing repo conventions. It emphasizes local UnitTest runs and warns that IntegrationTests depend on external services, which keeps indie-sized setups fast and reliable. The skill pairs naturally with the project-structure skill for navigation. It is not a general C# testing tutorial—it is the exact command cheat sheet Agent Framework maintainers expect before opening a PR.40installs4Python Package ManagementPython Package Management is an agent skill for Microsoft’s Agent Framework Python monorepo. Solo builders and small teams shipping custom connectors or patching framework packages use it when adding dependencies, cutting releases, or scaffolding new provider packages without breaking the shared core layout. It walks through the packages/ tree, how agent-framework-core relates to connector wheels, and how the root meta-package pulls optional extras. Operational commands center on Astral uv for lockfile-faithful installs and poethepoet tasks for repeatable setup, venv creation, and intentional pin upgrades. The lazy-loading pattern guidance matters when you want optional integrations to stay import-light. Reach for this skill during active framework development—not when you only consume the published SDK from an app repo.33installs5Python TestingPython Testing is an agent skill for contributors working in the Microsoft Agent Framework Python monorepo. It tells your coding agent how tests are organized, how to run them with uv and Poe, and what quality bar the maintainers expect—roughly 85% coverage on core paths, fast unit tests on every PR commit, and integration tests before merge. The guidance covers async pytest conventions (auto asyncio mode), marker-based filtering, per-package runs, and direct package directory execution when debugging. Solo builders shipping agents on top of Agent Framework benefit when they fork or patch packages locally; framework maintainers use it whenever creating or modifying tests. Invoke it whenever you add features, fix bugs, or refactor Python under `packages/` so new code ships with matching unit and integration coverage instead of breaking the project’s staged test pipeline.30installs6Python Samplespython-samples is a procedural skill for solo and indie contributors working in Microsoft’s Agent Framework Python tree who need new or updated demos to look like first-class citizens. It encodes the exact file skeleton teams expect—from optional PEP 723 inline script blocks and the Microsoft copyright line through asyncio-backed main entry points—so agents do not improvise structure. It also separates sample-only packages from the monorepo dev environment, which prevents dependency drift that breaks other contributors. The skill pushes over-documentation: README per sample folder, narrative module docstrings, and numbered walkthrough comments so readers can follow without spelunking the framework. When you wire an agent to extend the sample gallery, invoke this before generating files so formatting, pyright, and poe syntax checks pass on the first review.27installs7Project StructureProject-structure is an orientation skill for the Microsoft Agent Framework .NET solution. It does not generate code; it explains how the repository is laid out so solo builders and small teams can navigate dozens of related packages without guessing where abstractions, providers, and workflow orchestration live. The SKILL.md tree under dotnet/src lists first-party packages from core agents through OpenAI, Anthropic, Foundry, and Agent-to-Agent providers, plus workflows and legacy Azure integrations. It clarifies that tests follow predictable naming beside each project and that samples demonstrate end-to-end usage patterns. Shared and LegacySupport folders are called out as cross-project include targets with their own README guidance. Use this skill at the start of any agent-framework contribution session—implementing a provider tweak, adding a sample, or debugging integration tests—so the agent cites the correct project boundary instead of searching the whole repo blindly.26installs