What does this skill do, and when should you use it?
This skill supports JavaScript and TypeScript development with the Azure AI Projects SDK. It covers Foundry project clients, agents, connections, deployments, datasets, indexes, evaluators, and memory stores. It also explains how to obtain an OpenAI client, create conversations, run versioned agents, and clean up resources. It is a good fit for teams already using Azure AI Foundry or integrating its project resources into a TypeScript application.
It shows how to install @azure/ai-projects and @azure/identity, configure the project endpoint, model deployment name, and credentials, and use AIProjectClient operation groups. The examples create agents with code interpreter, file search, web search, Azure AI Search, function, and MCP tools. They also demonstrate retrieving an OpenAI client, creating responses and conversations, listing or retrieving connections and deployments, uploading datasets, managing indexes, and deleting resources.
- A TypeScript developer building an Azure AI Foundry application that needs versioned agents.
- An engineering team that must inspect project connections, credentials, and model deployments.
- An application developer uploading dataset files, managing dataset versions, or creating Azure AI Search indexes.
- A developer using Azure AI Projects together with the OpenAI responses or conversations APIs.
- A team adding code interpreter, file search, web search, function, or MCP tools to Foundry agents.
How do you install this skill?
- Treat API keys, SAS tokens, and connection metadata returned by getWithCredentials() as sensitive; confirm targets, scope, and cost before creation, upload, evaluation, and deletion operations.
- Verify createVersion, evaluations, memoryStores, and OpenAI responses/conversations APIs against the target SDK version and official documentation before use; the supplied material does not prove direct compilation or execution.
- Add documented timeout, retry, throttling, pagination, authorization, missing-resource, and partial-failure diagnostics and recovery strategies.
- Shell / CLI
- Network access
- Local filesystem
Node.js@azure/ai-projects@azure/identity
Install the collection into the chosen coding agent's skill directory:
npx skills add microsoft/skills
Select azure-ai-projects-ts in the wizard. For manual installation, clone https://github.com/microsoft/skills and copy .github/plugins/azure-sdk-typescript/skills/azure-ai-projects-ts/. Install the runtime packages with:
npm install @azure/ai-projects @azure/identity
For tracing, also install:
npm install @azure/monitor-opentelemetry @opentelemetry/api
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/microsoft/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/.github/plugins/azure-sdk-typescript/skills/azure-ai-projects-ts" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Use @azure/ai-projects to create a versioned TypeScript Azure AI Foundry agent with file search and authenticate with DefaultAzureCredential.
In an Agent Skills-compatible coding agent, try: “Use @azure/ai-projects to create a versioned TypeScript Azure AI Foundry agent with file search and authenticate with DefaultAzureCredential.” Set AZURE_AI_PROJECT_ENDPOINT and MODEL_DEPLOYMENT_NAME, and configure AZURE_TOKEN_CREDENTIALS when using the documented production credential pattern. The SKILL.md does not define a separate CLI entry point or executable script.
What are this skill's strengths and limitations?
- Covers the principal Azure AI Projects SDK operation groups with copyable TypeScript examples.
- Documents DefaultAzureCredential, ManagedIdentityCredential, project connections, and credential retrieval.
- Includes several agent-tool configurations plus OpenAI responses and conversations usage.
- The repository is MIT-licensed and supports collection installation through npx skills add.
- Its scope is the TypeScript @azure/ai-projects SDK and should not be treated as guidance for sibling Azure SDK skills.
- Examples require an Azure AI project endpoint, model deployment, and appropriate credentials.
- Some examples assume existing connections, deployments, vector stores, or indexes; provisioning workflows are not documented here.
- The README describes the repository as under active development, so content and testing may change.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Azure AI Projects for TypeScript this page ✓ Microsoft · Official | 49 · Use with care | ★ 2.7k | 3mo ago | MIT |
| Azure AI Search for TypeScript ✓ Microsoft · Official | 50 · Use with care | ★ 2.7k | 3mo ago | MIT |
| Azure Identity for TypeScript ✓ Microsoft · Official | 57 · Use with care | ★ 2.7k | 3mo ago | MIT |
| Azure Blob Storage for TypeScript ✓ Microsoft · Official | 51 · Use with care | ★ 2.7k | 3mo ago | MIT |
| Azure Document Intelligence for TypeScript ✓ Microsoft · Official | 50 · Use with care | ★ 2.7k | 3mo ago | MIT |
How did FollowSkills review this skill?
The documentation recommends DefaultAzureCredential, AAD, and avoiding credential logging, and the MCP example sets require_approval: "always", showing partial least-privilege and credential-safety awareness. Points are deducted because credential retrieval, dataset uploads, and agent/index/evaluation creation or deletion lack user confirmation, permission scoping, data-flow disclosure, and rollback or recovery guidance; secret handling for environment variables and endpoints is also underspecified.
Installation, authentication, operation groups, and some 404/403 handling are documented, and the connections reference includes an abnormal-path example. Points are deducted because there is no pinned version, compatibility matrix, timeout/retry/pagination/rate-limit guidance, or complete failure handling; several examples rely on undeclared helpers or context, and the supplied static material contains no executable reproduction evidence for this skill's key paths.
The audience and scenarios are reasonably clear, covering Foundry project clients, agents, connections, deployments, datasets, indexes, and evaluations. Points are deducted because non-fit boundaries, preview/service status, required Azure resources and permissions, and regional availability are not stated, and there is no Chinese-language guidance. No extra major deduction is made for mainland-China reachability because the evidence does not show total dependence on unreachable overseas services.
The skill has front matter, installation and environment-variable sections, progressive topic-based references, best practices, an MIT license, and author/version metadata. Points are deducted because package and API versions are not pinned, and there is no changelog, explicit maintenance owner or update path, FAQ, systematic limitations section, or dependency-security guidance; reference links and coverage are also somewhat inconsistent.
Examples cover the core workflow from client initialization through agents, connections, datasets, indexes, and evaluations, providing some directly reusable value. Points are deducted because the code was not validated beyond static inspection, several examples omit required context or helpers such as getAccessToken, vectorStoreId, and testData, and direct compilation or runtime compatibility with the current SDK is unverified; therefore the static ceiling limits this score.
The material contains auditable TypeScript examples, interface shapes, error-handling snippets, and references to official Microsoft documentation, while the repository includes general CI and test-harness material. Points are deducted because the supplied tests do not cover this skill's key paths, there is no skill-specific acceptance result, independent corroboration, or pinned dependency evidence, and no code was executed for this review.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →