microsoft-foundry

Deploy and test Azure AI Foundry agents by orchestrating project onboarding, agent creation, invocation testing, and troubleshooting workflows. The process uses mandatory Foundry MCP tool discovery and environment resolution from agent-metadata files to ensure safe, repeatable operations.

Eval-driven prompt optimization works by running batch evaluations, analyzing trace-to-dataset pipelines, comparing versions, and refining instructions. It uses strict eval/trace parameter conventions to identify quality gaps and optionally enable continuous production monitoring.

Yes, Azure AI Foundry agent invocation requires minimum RBAC permissions and capacity planning. The deployment process routes model decisions through quota and capacity discovery to ensure the necessary access rights and infrastructure are configured.

The best way to troubleshoot Azure AI Foundry agent failures is by using structured sub-skills for each workflow, reading mandatory sub-skill documents, and applying structured cache and metadata conventions to isolate and resolve issues across hosted and prompt agents.

Yes, you can set up continuous monitoring for Foundry agents by enabling continuous production evaluation and utilizing evaluator management. This integrates with trace-to-dataset evaluation pipelines to maintain agent quality over time.