microsoft / foundry-ai-solution-templates-creation

It supports 8 project types — RAG chatbot, multi-agent system, API backend, data pipeline, Azure Functions, full-stack web app, ML training/inference, and event-driven microservices — each backed by its own reference file with curated questions, folder structure, source patterns, Bicep modules, and a quality checklist. The skill asks 11 universal questions plus 6 to 11 type-specific ones, then generates the source code, Azure Bicep infrastructure, azure.yaml, Docker configuration, GitHub Actions CI/CD, a README with architecture diagrams, and Responsible AI documentation for whichever type was chosen.

This is for a developer or team that has already decided to build on Azure and wants a working project skeleton for one of the 8 supported patterns instead of assembling Bicep, azd configuration, and CI by hand. The multi-agent type always uses Microsoft Foundry and MAF, the ML-training type uses Azure ML instead, and six of the remaining types offer Foundry Agent Service as an optional add-on; a project that doesn't fit any of the 8 patterns falls outside what this skill is built to produce.

READ THE FULL ANALYSIS＋

What "ready" means here. Our dependency label for this skill is ready — no account or local tool is required before it can run its interview and write files. That covers only the scaffolding step itself: the project it hands you still needs azd and an Azure subscription to actually deploy, and several of the 8 types assume specific paid Azure services, such as AI Search, Cosmos DB, or AI Foundry, that the generated Bicep will attempt to provision.

Our capture stops partway through the README. The document is 28KB and we only read the first 12KB, which ends mid-way through the adaptive-questioning section — its Known Limitations and unsupported-project-type fallback sections are past that cutoff and aren't reflected in this summary.