// the find
Azure-Samples/aspire-semantic-kernel-creative-writer
A creative writing multi-agent solution to help users write articles using Aspire and Semantic Kernel
A .NET port of Microsoft's contoso-creative-writer sample: a multi-agent writing pipeline (researcher, writer, editor, marketing agents) built on Semantic Kernel's Agent framework and orchestrated locally via .NET Aspire, with Bing Search and Azure AI Search for grounding. Aimed at .NET developers who want a concrete reference for wiring Semantic Kernel agents, Aspire, and Azure AI Foundry together rather than reading the Python original.
Agent prompts are defined as separate yaml files per role (researcher/writer/editor/marketing), which is a cleaner separation than burying prompts in C# strings. It ships actual LLM-judged evaluation tooling (aieval report, scored on Coherence/Fluency/Relevance/Groundedness against three real scenarios in eval_inputs.json) instead of just claiming the app works. The two .NET Interactive notebooks under experiments/ let you poke at the WriterReviewer and full multi-agent flow in isolation before dealing with the API/React/Aspire stack.
Everything is pinned to preview surfaces: Azure AI Agent Service and the Semantic Kernel Agent SDK are both explicitly called out as preview, and gpt-4o/text-embedding-3-large are restricted to a couple of regions (eastus2, swedencentral) - this is brittle ground to build on. There are no unit or integration tests for the actual agent orchestration logic in ChatApp.WebApi; the only quality signal is the LLM-graded eval harness, which doesn't catch logic bugs. The README's own security notice says not to put this in production without more work, and at 66 stars for something this elaborate it hasn't seen much outside usage to shake out the rough edges.