Nabs.Launchpad.Core.Ai.Harness.Abstractions 10.0.249

Nabs.Launchpad.Core.Ai.Harness.Abstractions

Nabs.Launchpad.Core.Ai.Harness.Abstractions provides reusable console harness primitives for AI agents, including reactive terminal UI, command handling, observer pipelines, follow-up flows, and tracing helpers.

All the code in the assembly was copied from: https://github.com/microsoft/agent-framework/tree/main/dotnet/samples/02-agents/Harness

I may add additional features or alter this code over time.

Installation

dotnet add package Nabs.Launchpad.Core.Ai.Harness.Abstractions

The Problem

// 1) Every agent console app re-implements input loops, rendering, and shutdown behavior
while (true)
{
	var text = Console.ReadLine();
	if (text == "/exit") break;
	// duplicated orchestration logic
}

// 2) Streaming, errors, and tool output are mixed together in ad-hoc formatting code
await foreach (var update in agent.RunStreamingAsync(messages, session, options))
{
	Console.Write(update.Text); // no observer separation, hard to evolve
}

// 3) Follow-up questions require brittle state management
// pendingQuestions, responses, mode transitions all tracked manually.

// 4) No consistent command model (/exit, /todo, /mode, /session)
// each app parses commands differently and drifts over time.

Features

• Shared harness entry point via HarnessConsole.RunAgentAsync.
• Reactive console UI using HarnessAppComponent and UX state drivers.
• Composable observer pipeline for text, reasoning, tool calls, usage, and errors.
• Built-in command handlers for exit, todo, mode, and session workflows.
• Follow-up question orchestration with queueing and continuation support.
• OpenTelemetry tracing helper with file export through HarnessTracing.CreateFileTracerProvider.

Usage

using Microsoft.Agents.AI;
using Nabs.Launchpad.Core.Ai.Harness.Abstractions.SharedConsole;

AIAgent agent = BuildAgent();

await HarnessConsole.RunAgentAsync(
	agent,
	userPrompt: "Ask the agent anything",
	options: new HarnessConsoleOptions
	{
		Observers = HarnessConsoleOptions.BuildDefaultObservers(),
		CommandHandlers = HarnessConsoleOptions.BuildDefaultCommandHandlers(agent)
	});

using Microsoft.Agents.AI;
using Nabs.Launchpad.Core.Ai.Harness.Abstractions.SharedConsole;

AIAgent agent = BuildPlanningAgent();

var options = new HarnessConsoleOptions
{
	Observers = HarnessConsoleOptions.BuildObserversWithPlanning(
		agent,
		planModeName: "plan",
		executionModeName: "execute"),
	CommandHandlers = HarnessConsoleOptions.BuildDefaultCommandHandlers(agent)
};

await HarnessConsole.RunAgentAsync(agent, "Plan and execute tasks", options);

Configuration

• Use HarnessConsoleOptions.Observers to provide custom observer sets or disable observers with an empty list.
• Use HarnessConsoleOptions.CommandHandlers to customize command behavior before user input reaches the agent.
• Configure mode colors through HarnessConsoleOptions.ModeColors.
• Override session creation with HarnessConsoleOptions.SessionFactory when you need custom session lifecycle behavior.

Project Structure

• SharedConsole/HarnessConsole - high-level entry point that wires UI component and runner.
• SharedConsole/HarnessConsoleOptions - observer, command, mode color, and session configuration.
• SharedConsole/HarnessAgentRunner - orchestrates agent turns, streaming updates, commands, and follow-up flows.
• SharedConsole/HarnessAppComponent - reactive terminal component with input, list selection, and streaming panels.
• SharedConsole/Observers/* - pluggable observers for text, reasoning, usage, errors, tool calls, and planning output.
• SharedConsole/Commands/* - built-in slash-command handlers.
• SharedConsole/HarnessTracing - OpenTelemetry tracing provider setup with file span export.

License

Copyright (c) Net Advantage Business Solutions.