Azure AI Foundry Projects for .NET
Guides .NET developers through Foundry project, agent, and evaluation operations.
The document specifies DefaultAzureCredential, environment variables, cleanup, and limiting credential inclusion to cases that need it. It does not define least-privilege Azure roles, user confirmation, sensitive-data handling, external-tool data flows, recovery for deletion operations, or dependency security review, so points are deducted.
Installation, authentication, client hierarchy, and several workflows are broadly self-consistent, with repository-level CI and test signals. However, no skill-specific tests, reproducible verification, comprehensive abnormal-input handling, or complete pagination/error behavior is shown, and GA and preview APIs are mixed; the static cap keeps this below 10.
The name, .NET audience, SDK scope, and trigger terms are reasonably clear, covering agents, connections, datasets, deployments, evaluations, and indexes. Non-fit scenarios, input/output boundaries, version-selection rules, Chinese-language support, and mainland-China reachability are not specified, so points are deducted.
The skill includes installation, environment variables, authentication, workflow sections, type references, best practices, error handling, related SDKs, and version information. MIT licensing and Microsoft provenance are clear, but changelog, explicit maintenance commitments, FAQs, known limitations, and fuller troubleshooting are missing, so points are deducted.
It supplies directly adaptable C# examples for several common Foundry project operations and communicates a clear core benefit. There is no skill-specific build or runtime evidence in the supplied files, some calls may be sensitive to SDK version or preview status, and comparative benefit or production completeness is limited, so points are deducted.
Package versions, API references, source links, samples links, and repository-level CI/test materials provide some traceability. The target skill lacks dedicated acceptance criteria, a covering test suite, or third-party execution evidence, and this review is static only, so evidence coverage remains limited.
- The examples mix GA, preview, and --prerelease packages; verify SDK versions, API signatures, and compatibility before use.
- includeCredentials:true, dataset uploads, index or dataset deletion, and external tool calls may expose sensitive data or create persistent external effects; confirmation, permission, and recovery procedures are not documented.
- No skill-specific acceptance tests or runtime results are supplied, so compilation, complete service configuration, and end-to-end workflow usability are unconfirmed.
- Chinese-language usability and reachability of Azure, NuGet, documentation, and Copilot-related services from mainland-China networks are not addressed.
What it does & when to use it
This skill provides .NET development guidance for Azure AI Foundry through the Azure.AI.Projects SDK. It covers project-level operations for agents, connections, datasets, deployments, evaluations, and indexes, with additional examples for persistent and preview versioned agents. It explains configuration for the project endpoint, model deployment, and DefaultAzureCredential. It fits developers building or maintaining Azure AI Foundry applications in .NET.
Shows how to install Azure.AI.Projects, Azure.Identity, and optional extension packages; create an AIProjectClient with a project endpoint and credential; create threads, messages, and runs, poll persistent-agent status, and read messages; create versioned agents with tools and obtain responses; list and retrieve connections, deployments, datasets, indexes, and evaluations; upload or delete file and folder datasets; create Azure AI Search indexes; run evaluations; and use an Azure OpenAI ChatClient for chat completion. It also demonstrates RequestFailedException handling and resource cleanup.
- A C# developer managing Foundry project connections, model deployments, datasets, and indexes.
- A .NET team building persistent agents with threads, messages, run polling, message retrieval, and cleanup.
- A team creating preview versioned agents with the Web Search tool and calling them through a project response client.
- An engineer configuring and running relevance evaluations against a project dataset.
- An application developer obtaining an Azure OpenAI client from a Foundry project connection and performing chat completion.
Pros & cons
- Covers major AIProjectClient areas, including agents, connections, datasets, deployments, evaluations, and indexes.
- Provides copyable C# examples for persistent agents, versioned agents, Azure AI Search, and evaluations.
- States the GA v1.1.0 and preview v1.2.0-beta.5 versions and identifies the related NuGet packages.
- Includes guidance on authentication, polling, pagination, credential access, and resource cleanup.
- Focused on .NET and Azure AI Foundry, so it does not replace skills for other language SDKs.
- Versioned-agent and persistent-agent extensions use prerelease packages and preview APIs.
- Requires an Azure AI project endpoint, model deployment, and suitable credentials; the source does not provide a complete permissions matrix.
- The source does not provide independent test coverage or failure-case documentation for this individual skill.
How to install
Add the collection in Copilot CLI with npx skills add microsoft/skills, then select this skill in the installation wizard. The README says skills are installed into the selected agent directory, such as .github/skills/ for GitHub Copilot. This skill is located at .github/plugins/azure-sdk-dotnet/skills/azure-ai-projects-dotnet/SKILL.md. Install the main SDK dependencies with dotnet add package Azure.AI.Projects and dotnet add package Azure.Identity. For versioned agents, optionally run dotnet add package Azure.AI.Projects.OpenAI --prerelease; for low-level agent operations, optionally run dotnet add package Azure.AI.Agents.Persistent --prerelease.
How to use
Give the coding Agent a task containing the relevant intent, such as: “Use AIProjectClient to connect to a Foundry project, list deployments, and run an evaluation” or “Create a .NET versioned agent and call its Web Search tool.” Before running generated code, set PROJECT_ENDPOINT and MODEL_DEPLOYMENT_NAME; set CONNECTION_NAME or AI_SEARCH_CONNECTION_NAME when using project connections. For production use of DefaultAzureCredential, configure AZURE_TOKEN_CREDENTIALS=prod or a specific credential name as shown. The README does not document a separate test command or standalone package for this individual skill.
Compared to similar skills
Compared with Azure.AI.Agents.Persistent, this skill centers on the higher-level Azure.AI.Projects project client while exposing low-level agent operations through GetPersistentAgentsClient(). Azure.AI.Projects.OpenAI is presented as an optional preview extension for versioned agents and OpenAI integrations.