MultiAgentSemanticKernel

A .NET 9 console demo app showcasing Semantic Kernel multi‑agent orchestration patterns end‑to‑end. It wires up DI, options, Azure OpenAI via DefaultAzureCredential, ANSI console output, and multiple orchestration styles (sequential, concurrent, group chat, handoff, magentic).

Requirements

• .NET SDK 9
• Azure OpenAI access (for real runs)
• Authentication via DefaultAzureCredential (e.g., Azure CLI login, Managed Identity, or Visual Studio/VS Code sign-in)

Setup

# Restore and build
dotnet restore
dotnet build

Configuration

Configuration uses the Options pattern bound from appsettings.json and environment variables (prefix MASKE_).

• File: appsettings.json

{
  "AzureOpenAI": {
    "Endpoint": "https://YOUR-AOAI-ENDPOINT.openai.azure.com/",
    "Deployments": {
      "Llm": "your-llm-deployment"
    }
  },
  "EnableAgentLogging": false
}

• Environment overrides (prefix MASKE_):
  • MASKE_AzureOpenAI__Endpoint
  • MASKE_AzureOpenAI__Deployments__Llm
  • MASKE_EnableAgentLogging (bool; if true, agent/orchestration logging is wired into SK)

The kernel is configured with Azure OpenAI chat completion using the provided endpoint and deployments. Credentials are sourced from DefaultAzureCredential.

Running

First argument selects the runner mode. Remaining args are treated as the prompt. If no prompt is supplied, each runner uses a sensible baked‑in default for demo purposes.

Usage:
  dotnet run -- <Sequential|Concurrent|GroupChat|Handoff|Magentic> [prompt...]

# Examples
dotnet run -- Sequential "As a user, I can upload avatars up to 2MB."
dotnet run -- Concurrent "Analyze PR: feat(auth): add input validation and fix null handling"
dotnet run -- GroupChat "Move session state to Azure Cache for Redis Enterprise, SKU E3"
dotnet run -- Handoff "Add dark mode feature toggle and roll it out safely"
dotnet run -- Magentic "Stabilize error budget for service 'catalog'"

Runners

• Sequential: Executes a deliberate pipeline of agents (BacklogRefiner, Scaffolder, APIDesigner, TestWriter, DocWriter) using Sequential Orchestration. Imports DevWorkflowPlugin.
• Concurrent: Runs multiple analysis agents concurrently (DiffAnalyst, TestImpactor, SecLint, Compliance). Imports PrAnalysisPlugin.
• GroupChat: Round‑robin group chat between TechLead, SRE, Security, and DataEng.
• Handoff: Triage, Design, and Implementation agents using explicit handoff rules and an interactive callback.
• Magentic: Ops‑focused flow (DeployInspector, Deployer, Notifier) using Magentic Manager with tools from OpsPlugin.

Console UX

Console output is optimized for readability:

• Agent lifecycle and tool calls are rendered with ANSI colors and icons.
• User input is highlighted via a dedicated UserInput block for dark‑mode friendly contrast.
• Results are printed compactly after orchestration completes.

What’s implemented

• DI via Host.CreateApplicationBuilder
• Options model in Options/AzureOpenAIOptions.cs
• Semantic Kernel 1.65.0 with Azure OpenAI chat completion
• SK Agents packages (preview) for orchestration patterns
• Authentication via DefaultAzureCredential
• Console logging, plus a function invocation filter (ConsoleFunctionInvocationFilter) to display tool invocations
• Plugins: DevWorkflowPlugin, PrAnalysisPlugin, OpsPlugin (+ OpsInspectorTools, OpsDeployerTools, OpsNotifierTools)

Notes

• If you don’t have valid Azure OpenAI configuration, calls will fail at runtime.
• Some runners use longer timeouts (e.g., 120–300s) to await model output.