← Back to Talks
January 24, 2026 Advisory Hour Episode 1 Agent Skills

I Built My First Agent Skill in VS Code

Claude Skills, Copilot Skills, and the messy reality of building with AI dev tools — featuring real-time troubleshooting, MCP conflicts, and a book about embarrassing yourself in public.

🎯 The Premise

I'm starting something new: Advisory Hour — where I combine books I'm reading with emerging AI dev tools to build practical workflows, showing the thinking process along the way. First episode: building a simple Agent Skill inside VS Code using the Agent Skills folder structure.

What You'll See

Tools & Concepts

Claude Skills GitHub Copilot VS Code Agent Workflows MCP Tools PowerShell Python Prompt Engineering

Key Discovery: Skills vs MCP Conflicts

One of the most interesting moments was discovering what happens when your new skill conflicts with an existing MCP tool. I had a telnet MCP server set up for MUD games, and the agent kept trying to route RPG character generation through it.

🔍 The Conflict

When you create a skills file but have an MCP with similar functionality, the agent can get confused about which tool to use. This required explicitly switching to the correct model and re-prompting to get the agent to recognize the local skill vs the remote MCP tool.

The Workflow Evolution

What started as a simple Python script turned into a multi-step evolution:

📚

Book Referenced

Public Speaking with Confidence (Kindle Unlimited) — early chapter on intentionally embarrassing yourself to build comfort and confidence. The core idea: commit to embarrassing yourself in a controlled way to overcome fear of public judgment.

What This Represents

This video captures something important about the current moment in AI tooling: the gap between "AI influencers talking about AI" and "real-world problems that need solving."

I'm trying to reconcile two worlds: the AI-native world where everything accelerates, and the skeptical world where people ask "what benefit does this actually bring?" The answer isn't always clear, but the process of building — even when messy — reveals where the technology actually helps versus where it's just another toy demo.

💡 Bottom Line

Skills can find generated code and solve problems, but that hasn't materially made my coffee better. The question isn't "can AI do this?" but "what brings material benefit versus novelty?"

📝 Transcript Excerpt

"What we've learned: if you have a skills file and you have existing tools, you may see some conflict between those two. Need to untangle how to make your skill files work with your tools correctly... The system is more than capable of bootstrapping itself and correcting errors as you go. And then the next thing you know, you have an RPG character generator inside it."

Watch on YouTube

This is the first episode of Advisory Hour. If you find it valuable, like and subscribe on YouTube to help me gauge whether to keep going.

Watch on YouTube →