Azure AI Persistent Agents for Java
Guides Java developers through building and managing Azure AI agents with persistent threads, messages, runs, and tools.
The skill recommends DefaultAzureCredential and places endpoint, model, and credential configuration in environment variables. No credential theft, covert exfiltration, malware, or other red-line risk is visible. However, creating, uploading, running, and deleting cloud resources has external effects, while least-privilege guidance, user confirmation, data-flow disclosure, sensitive-file handling, rollback, and pre-action checks are absent, so points are deducted.
Installation, authentication, sync/async clients, threads, messages, runs, streaming, and error paths are documented, including several run states. However, there is no skill-specific execution or test evidence, polling lacks timeout and cancellation controls, tool-call handling is incomplete, and compatibility with the beta dependency is unverified. Static calibration therefore limits the score and further deductions apply.
The trigger phrases, Java SDK target, and main scenarios are reasonably clear, covering persistent agents, threads, messages, runs, streaming, and tools. Boundaries, non-fit cases, input/output contracts, false-trigger conditions, and Chinese-language guidance are not defined, and use assumes an Azure project, model deployment, and reachable cloud environment, so points are deducted.
The documentation is readable and layered, with installation, environment variables, authentication, workflow, best practices, error handling, and references. It includes MIT licensing, author, and version metadata. However, the dependency is beta.1 despite the 1.0.0 metadata, and there is no skill-specific changelog, maintenance owner, update path, known-limitations section, FAQ, or complete prerequisite documentation, so points are deducted.
The examples cover many common SDK operations and provide useful implementation patterns with clear marginal value over starting manually. They are not statically verified, may require missing imports or version adjustments, and leave polling, tool execution, and response selection for engineering work. The result is therefore a useful starting point rather than directly deliverable output, so points are deducted.
Maven, GitHub source, and official documentation links are provided, and the repository shows generic CI and harness infrastructure. However, no skill-specific tests, acceptance results, or third-party execution evidence are supplied, so compilation and runtime behavior cannot be confirmed from the static material and only limited credit is justified.
- Verify that azure-ai-agents-persistent beta.1 matches the documented APIs and add skill-specific compile and integration tests before relying on the examples.
- Before creating, uploading, running, or deleting cloud resources or files, clarify authorization scope, data flow, file sensitivity, cost impact, and recovery options.
- Add polling timeouts, cancellation, retry/backoff behavior, and complete abnormal-state handling; do not execute high-impact external tools without confirmation.
- Chinese guidance and mainland-China network reachability are not established; users must verify access to Azure services, dependency repositories, and documentation.
What does this skill do, and when should you use it?
This skill is for the Azure AI Agents Persistent SDK for Java and focuses on its low-level API. It covers client construction, authentication, agent and thread lifecycle, message creation, run polling, response retrieval, and cleanup. It also explains sync versus async clients, run-status handling, and HTTP error capture. It fits Azure Java projects that need direct control over persistent agent workflows.
Guides a coding agent to add the Maven dependency, read PROJECT_ENDPOINT, MODEL_DEPLOYMENT_NAME, and Azure credential environment variables, and build a PersistentAgentsClient. It demonstrates creating agents and threads, adding user messages, starting runs, polling every 500 milliseconds, listing responses, handling RequiresAction, Failed, and Cancelled statuses, and deleting threads and agents.
- A Java developer needs a persistent agent that can be reused across sessions in an Azure AI project.
- A backend engineer needs to implement a workflow built from threads, messages, and agent runs.
- A team needs to authenticate with DefaultAzureCredential or managed identity.
- A high-concurrency Java service needs to evaluate PersistentAgentsAsyncClient for improved throughput.
- A developer needs run polling, failure handling, and resource cleanup around agent operations.
What are this skill's strengths and limitations?
- Covers the full low-level workflow from agent creation through cleanup.
- Documents both synchronous and asynchronous clients.
- Includes authentication, run-status, and HTTP error-handling examples.
- Provides copyable Maven and Java snippets.
- It is low-level SDK guidance rather than a high-level framework or complete application template.
- The source does not specify a Java version, build configuration details, or a test suite.
- The examples use version 1.0.0-beta.1, a beta release.
- Users must provide the Azure project endpoint, model deployment, and credentials.
How do you install this skill?
Install the collection with npx skills add microsoft/skills, then select the required skill in the wizard. Add this Maven dependency to the Java project: <dependency><groupId>com.azure</groupId><artifactId>azure-ai-agents-persistent</artifactId><version>1.0.0-beta.1</version></dependency>. The source does not document a command for selecting this exact skill path directly.
How do you use this skill?
After installation, prompt the coding agent: “Use PersistentAgentsClient to create a Java persistent agent with a thread, user message, run polling, and response retrieval.” Supported trigger phrases include “PersistentAgentsClient”, “persistent agents java”, “agent threads java”, “agent runs java”, and “streaming agents java”. Set PROJECT_ENDPOINT and MODEL_DEPLOYMENT_NAME; when using DefaultAzureCredential, also set AZURE_TOKEN_CREDENTIALS=prod.