Microsoft Foundry Python Projects Skill
Gives coding agents dependable patterns for building Foundry AI applications with the Azure AI Projects Python SDK.
The skill prefers DefaultAzureCredential, constrains production credential chains, and emphasizes deterministic client and credential cleanup, providing some least-privilege and audit guidance. However, it also covers API keys, no-auth credentials, MCP, OpenAPI, computer-use, and external connections; its include_credentials example retrieves credentials, while per-operation confirmation, data-flow disclosure, sensitive-data handling, permission boundaries, and rollback guidance are missing. Eleven points are deducted.
Installation, sync/async separation, some error handling, and cleanup guidance are present, so the main paths are plausible from static inspection. There are no skill-specific tests, several examples show possible API or version inconsistencies such as McpTool versus MCPTool and differing tool import paths, async credential management is inconsistent across snippets, and the supplied API reference is incomplete. Fourteen points are deducted; the static ceiling also prevents a higher score.
The name, audience, and high-level Foundry SDK positioning are clear, and the skill identifies the lower-level agent SDK as an alternative. It covers sync, async, agents, evaluation, connections, datasets, and indexes. However, precise trigger conditions, explicit non-fit boundaries, input/output contracts, Chinese-language support, mainland-China reachability, and regional service constraints are not documented. Seven points are deducted.
The skill provides YAML metadata, an MIT license, a version, installation instructions, organized reference files, and many examples. It lacks a changelog, explicit maintenance ownership and update path, and stronger management of beta-version/API drift; referenced scripts and some reference files are not present in the supplied evidence. Six points are deducted.
The broad coverage can provide directly adaptable code scaffolding for common Foundry Python SDK tasks, including lifecycle, evaluation, and tool patterns. There are no verified representative outputs, end-to-end demonstrations, or comparisons with official alternatives, and some snippets may require substantial manual validation because of SDK-version or interface inconsistencies. Six points are awarded under the static ceiling, with nine points deducted.
The API reference identifies a package version, upstream repository, and commit, while several references link to official documentation or samples, giving partial traceability. There is no skill-specific test suite, CI coverage, execution log, or independent reproduction evidence, and the supplied API reference is incomplete. Six points are deducted.
- Before use, reconcile every API example with the installed azure-ai-projects version and add executable tests; the supplied material does not establish that the snippets run.
- Do not use include_credentials=True or print credential-related fields in production; define explicit scopes and user confirmation for MCP, OpenAPI, computer-use, file uploads, and external connections.
- The skill is English-only and does not address mainland-China network reachability, Azure regional availability, or service dependency constraints.
What does this skill do, and when should you use it?
This is a Python skill in Microsoft’s Agent Skills monorepo for the azure-ai-projects SDK used with Microsoft Foundry. It guides agents through AIProjectClient usage, agent creation and versioning, connections, deployments, datasets, indexes, and evaluations. It covers both the native Foundry client and an OpenAI-compatible client, along with synchronous, asynchronous, and memory-store patterns. It is a good fit for teams already using Azure Foundry projects that want coding agents to produce consistent SDK code.
It instructs agents to install azure-ai-projects and azure-identity, read AZURE_AI_PROJECT_ENDPOINT and AZURE_AI_MODEL_DEPLOYMENT_NAME, and authenticate to a Foundry project with DefaultAzureCredential. It produces Python patterns for creating agents, PromptAgentDefinition versions, tools, threads, messages, runs, connections, deployments, datasets, indexes, evaluations, and memory stores. It also covers OpenAI-compatible calls and asynchronous clients.
- A Python developer using an Azure Foundry project needs to create a basic or tool-enabled agent.
- A production team needs DefaultAzureCredential, managed identity, and deterministic client cleanup.
- An engineering team needs versioned production agents built with PromptAgentDefinition and version labels.
- An engineer needs to manage model deployments, project connections, datasets, or indexes through the Foundry client.
- A developer needs to issue chat requests or run evaluations through an OpenAI-compatible client.
What are this skill's strengths and limitations?
- Covers the Foundry project client, agent versions, tools, evaluations, connections, deployments, datasets, indexes, and memory stores.
- Explicitly recommends DefaultAzureCredential and consistent sync or async resource management.
- Provides both native Foundry API patterns and an OpenAI-compatible client approach.
- The README records 12 test scenarios for this skill.
- It requires an Azure AI Projects endpoint and model deployment name and is not documented as a standalone offline skill.
- Examples depend on Azure identity authentication and network services.
- The repository README says the overall project is under active development, so the collection and patterns may change.
- The source does not specify pricing, quotas, or the exact permissions required by each Azure service.
How do you install this skill?
Install the collection with npx skills add microsoft/skills, then select the required skill in the wizard. For manual installation, clone https://github.com/microsoft/skills and place the target skill in the selected coding agent’s skill directory; the source does not provide a separate command for installing only azure-ai-projects-py. The Python packages used by the examples are installed with: pip install azure-ai-projects azure-identity.
How do you use this skill?
Ask a coding agent for an Azure AI Projects Python task, for example: “Create a versioned Foundry agent with AIProjectClient, DefaultAzureCredential, and PromptAgentDefinition, then run it through a thread.” Provide AZURE_AI_PROJECT_ENDPOINT and AZURE_AI_MODEL_DEPLOYMENT_NAME. In production, set AZURE_TOKEN_CREDENTIALS=prod or select a specific credential when using DefaultAzureCredential.
How does this skill compare with similar options?
The source contrasts this skill with the azure-ai-agents-python skill: azure-ai-projects is the higher-level Foundry SDK with versioning, connections, deployments, datasets, indexes, and evaluations, while azure-ai-agents-python is intended for lower-level standalone agent operations.