Microsoft Agent Framework
Agents
Created: 10 Jun 2026
Updated: 10 Jun 2026
Agent Skills
Adding Skills to Agents in the Microsoft AI Agent Framework
The Microsoft AI Agent Framework lets you add custom tools, policies, and expertise to agents via Skills. Skills use a progressive disclosure pattern so they stay low overhead until the agent specifically needs them.
In C#, there are three primary ways to define and provide these skills to your agents:
- File-based — Exposing a folder on the filesystem that contains a
SKILL.mdfile. - Code-defined — Using the
AgentInlineSkilldirectly within your source code. - Class-based — Encapsulating the logic, scripts, and resources within a class deriving from
AgentClassSkill<T>.
1. File-based Skills
File-based skills are loaded automatically from a folder structure. All you need is a SKILL.md file inside a folder. The system will inject the YAML frontmatter describing the skill so the agent knows when it might be relevant.
Full Example (SKILL.md)
---
name: car-rental-policy
description: Guidelines and policies for renting vehicles. Use when asked about mileage limits, insurance, or fuel policies.
---
# Car Rental Policy
- Insurance: Basic coverage is included. Premium is $15/day extra.
- Mileage: 200 miles per day free. $0.25 per additional mile.
- Fuel: Return with a full tank to avoid a $40 refuel fee.
Full Example (CarRentalFileSkillDemo.cs)
using System;
using System.IO;
using System.Threading.Tasks;
using Microsoft.Agents.AI;
using Azure.Identity;
using OpenAI;
using System.ClientModel;
using Azure.AI.OpenAI;
namespace MicrosoftAgentFrameworkLesson.ConsoleApp.Agents;
public class CarRentalFileSkillDemo
{
public static async Task RunAsync()
{
Console.WriteLine("--- Car Rental File-Based Skill Demo ---");
// The path to the parent directory containing skill folders
string skillPath = Path.Combine(AppContext.BaseDirectory, "Agents", "Skills");
// Provide the file-based skills
var skillsProvider = new AgentSkillsProvider(skillPath);
// Assume standard endpoint/credentials setup (skipped for brevity, but needed for completeness)
string endpoint = Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT") ?? "https://placeholder.openai.azure.com/";
var agent = new AzureOpenAIClient(new Uri(endpoint), new DefaultAzureCredential())
.GetChatClient("gpt-4o-mini")
.AsAIAgent(new ChatClientAgentOptions
{
Name = "CarRentalAgent",
ChatOptions = new()
{
Instructions = "You are a helpful car rental assistant.",
},
AIContextProviders = [skillsProvider] // Inject the skill!
});
Console.WriteLine("Agent configured with file-based skill successfully.");
await Task.CompletedTask;
}
}
2. Code-defined (Inline) Skills
Code-defined skills use the AgentInlineSkill class. This approach is highly dynamic and flexible, meaning you don't need external files to define resources or policy blocks.
Full Example (CarRentalInlineSkillDemo.cs)
using System;
using System.Threading.Tasks;
using Microsoft.Agents.AI;
using Azure.Identity;
using Azure.AI.OpenAI;
namespace MicrosoftAgentFrameworkLesson.ConsoleApp.Agents;
public class CarRentalInlineSkillDemo
{
public static async Task RunAsync()
{
Console.WriteLine("--- Car Rental Inline Skill Demo ---");
// Define a code-based skill inline
var pricingSkill = new AgentInlineSkill(
name: "car-pricing-skill",
description: "Calculates total cost for car rentals",
instructions: """
Use this skill to determine pricing for a requested car category.
1. Check the pricing-table resource for the daily rate.
2. Multiply the rate by the numbers of days requested.
""")
.AddResource(
"pricing-table",
"""
| Category | Daily Rate |
|----------|------------|
| Compact | $30 |
| Sedan | $45 |
| SUV | $65 |
""");
var skillsProvider = new AgentSkillsProvider(pricingSkill);
string endpoint = Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT") ?? "https://placeholder.openai.azure.com/";
var agent = new AzureOpenAIClient(new Uri(endpoint), new DefaultAzureCredential())
.GetChatClient("gpt-4o-mini")
.AsAIAgent(new ChatClientAgentOptions
{
Name = "CarRentalAgent",
ChatOptions = new() { Instructions = "You are a helpful car rental computing assistant." },
AIContextProviders = [skillsProvider]
});
Console.WriteLine("Agent configured with inline skill successfully.");
await Task.CompletedTask;
}
}
3. Class-based Skills
Using the AgentClassSkill<T> generic type allows developers to bind attributes to methods and properties easily, establishing clear structural ties directly within C# source files. Properties act as resources read by the agent, and methods become script executions triggered by the agent.
Full Example (CarRentalClassSkillDemo.cs)
using System;
using System.ComponentModel;
using System.Threading.Tasks;
using Microsoft.Agents.AI;
using Azure.Identity;
using Azure.AI.OpenAI;
namespace MicrosoftAgentFrameworkLesson.ConsoleApp.Agents;
internal sealed class CarRentalAvailabilitySkill : AgentClassSkill<CarRentalAvailabilitySkill>
{
public override AgentSkillFrontmatter Frontmatter { get; } = new(
"car-availability",
"Checks availability of specific cars in the rental lot. Use when users ask if a particular model is available.");
protected override string Instructions => """
Use this skill to check if a specific car or category is available.
1. Read the inventory-status resource to see what cars are ready.
""";
[AgentSkillResource("inventory-status")]
[Description("Current list of available cars to rent.")]
public string InventoryStatus => """
# Current Available Inventory
- 2x Toyota Corolla (Sedan)
- 1x Honda CR-V (SUV)
- 0x Ford Mustang (Compact) - OUT OF STOCK
""";
}
public class CarRentalClassSkillDemo
{
public static async Task RunAsync()
{
Console.WriteLine("--- Car Rental Class-based Skill Demo ---");
var skill = new CarRentalAvailabilitySkill();
var skillsProvider = new AgentSkillsProvider(skill);
string endpoint = Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT") ?? "https://placeholder.openai.azure.com/";
var agent = new AzureOpenAIClient(new Uri(endpoint), new DefaultAzureCredential())
.GetChatClient("gpt-4o-mini")
.AsAIAgent(new ChatClientAgentOptions
{
Name = "CarRentalAgent",
ChatOptions = new() { Instructions = "You are a car availability specialist." },
AIContextProviders = [skillsProvider]
});
Console.WriteLine("Agent configured with class-based skill successfully.");
await Task.CompletedTask;
}
}
Reference: Microsoft Agent Framework - Skills