A Skill is a self-contained, file-based instruction set that tells an AI agent how to handle a specific topic or task. Each skill lives in its own folder and is defined by a SKILL.md file.

skills/
└── my-skill/
    ├── SKILL.md          ← required: name, description, instructions
    ├── references/       ← optional: supporting docs the agent can read
    ├── assets/           ← optional: templates, forms
    └── scripts/          ← optional: executable scripts

The SKILL.md front-matter tells the agent when to activate the skill (description), and the body tells it what to do (instructions). The FileAgentSkillsProvider scans a directory, discovers all SKILL.md files, and injects the relevant skill into the agent's context at runtime — no code changes needed to add or update a skill.

A general-purpose interactive chat agent that demonstrates three different skill types: document lookup, Python script execution, and persona adoption.

1. FileAgentSkillsProvider loads all skills from TestData\AgentSkills\.
2. A PythonRunner tool (execute_python) is registered so the agent can run .py scripts.
3. ToolCallingMiddleware handles the tool-call loop automatically.
4. Utils.RunChatLoopWithSession starts an interactive console session.

Prerequisites: .NET 10, Python installed and on PATH, Azure OpenAI access.

1. Configure Azure OpenAI credentials via User Secrets:

   dotnet user-secrets set "AzureOpenAI:Endpoint" "https://<your-resource>.openai.azure.com/"
   dotnet user-secrets set "AzureOpenAI:ApiKey" "<your-api-key>"

2. Run the project:

   cd ConsoleApp
   dotnet run

3. Try these sample prompts:
   • What are the company leave benefits?
   • Run the secret formula script
   • Talk to me like a pirate

Create a folder under TestData\AgentSkills\ with a SKILL.md:

---
name: my-new-skill
description: Handles questions about <topic>
---

## Instructions
1. Step one...
2. Step two...

Mark the file as CopyToOutputDirectory in ConsoleApp.csproj — no code changes required.

A production-style IT HelpDesk agent for Contoso that handles the three most common employee IT requests using structured, policy-driven skills.

1. FileAgentSkillsProvider scans the skills/ directory at startup.
2. The agent is created with a system prompt that sets the Contoso HelpDesk persona.
3. Responses are streamed chunk-by-chunk for a responsive UX.
4. A persistent AgentSession maintains conversation context across turns.

Prerequisites: .NET 10, Azure OpenAI access.

1. Configure Azure OpenAI credentials via User Secrets or environment variables:

   dotnet user-secrets set "AzureOpenAI:Endpoint" "https://<your-resource>.openai.azure.com/"
   dotnet user-secrets set "AzureOpenAI:ApiKey" "<your-api-key>"

2. Run the project:

   cd ITHelpDeskAgent
   dotnet run

3. Try these sample prompts:
   • I've been locked out of my account — wrong password too many times.
   • My VPN keeps dropping on macOS, I've already tried restarting.
   • I need to install Postman for API testing, how do I request it?

Drop a new folder under skills/ — no code changes needed:

skills/
└── hardware-request/
    ├── SKILL.md
    └── references/
        └── HARDWARE_CATALOG.md

The skills\**\* glob in ITHelpDeskAgent.csproj copies everything automatically.

• Skill versioning — add version to SKILL.md front-matter and log which version handled each request for auditability.
• Fallback / escalation skill — a catch-all skill that creates a ServiceNow/Jira ticket when no other skill matches.
• Confidence threshold — reject or escalate responses below a confidence score to avoid hallucinated answers.

• Ticketing system integration — auto-create tickets in ServiceNow or Jira directly from the agent instead of instructing users to do it manually.
• Active Directory / LDAP lookup — verify employee identity and fetch account status before guiding through password reset.
• Email / Teams notifications — send confirmation emails or Teams messages after a software request is submitted.

• Role-based skill access — restrict sensitive skills (e.g., password-reset) to authenticated users only.
• Audit logging — persist every conversation turn with user ID, skill used, and timestamp to a database for compliance.
• PII redaction — strip employee IDs and emails from logs before storage.

• Skill caching — cache parsed SKILL.md content in memory so the file system is not hit on every request.
• Async Python execution — replace the synchronous Process.WaitForExit() in PythonRunner with async process handling to avoid thread blocking.
• Web API / Bot Framework host — expose the agent as an HTTP endpoint so it can be embedded in Teams, Slack, or a web portal instead of running as a console app.

• Skill unit tests — validate each SKILL.md parses correctly and the agent routes sample queries to the expected skill.
• Hot-reload skills — use a FileSystemWatcher to reload skills without restarting the application.
• Skill authoring CLI — a small tool to scaffold a new skill folder with the correct front-matter template.