Agent Skills: Four-Hour Masterclass

Design, build, and evaluate reusable skills that bring your expertise into Claude Code.

When you use an AI assistant for recurring work, you often need to explain the same procedures, supply the same reference material, and correct the same mistakes. Your expertise includes decisions about how to approach the task, which tools to use, and what a satisfactory result looks like.

In this four-hour, hands-on masterclass, you’ll learn how to capture that expertise in reusable Agent Skills. Using Claude Code as the main environment, you’ll build a skill around a recurring workflow, then extend it with supporting material, executable steps, and checks for output quality.

The session follows the development process from choosing a suitable task to testing, troubleshooting, and preparing the skill for reuse. Short explanations and live demonstrations alternate with guided exercises, giving you time to apply each concept and ask questions as your implementation develops.

                                               Sunday, 6 December 2026 · 7:00–11:00 pm EET (Finland time, UTC+2)
                                               Limited to 30 participants · $49 early bird through 20 November · $59 from 21 November

What you’ll build

You’ll build and test an Agent Skill for a recurring task of your choice—for example, reviewing technical documents, preparing a structured report, applying coding standards, or following a research procedure.

Your skill will define when it should be used, the process the agent should follow, the reference material it needs, and the output you expect. You’ll explore how to add scripts and external tools, then test both successful runs and cases where the workflow fails.

Bring a task you know well. Understanding the work and its quality standards will help you judge whether the skill performs it correctly.

What you’ll learn and how you can apply it

By the end of the masterclass, you’ll be able to:

• Choose the right approach: Decide when to use a skill, a standalone prompt, project instructions, or an MCP integration.
• Translate expertise into instructions: Define procedures, constraints, completion criteria, and expected outputs for a recurring task.
• Organize supporting material: Structure references and examples so the agent can use them during execution.
• Test skill activation: Write descriptions and check them against requests that should—and should not—activate the skill.
• Extend the workflow: Incorporate scripts and external tools while accounting for dependencies and permissions.
• Evaluate results: Identify meaningful quality checks, recognize misleading scores, and investigate failures systematically.
• Prepare for reuse: Package, document, share, and maintain skills for personal or team workflows.

Workshop agenda

1. Choose the workflow and define success

Examine how skills relate to prompts, project instructions, and MCP integrations. Identify a suitable recurring task, define its boundaries, and specify what the agent needs to produce.

Exercise: Write a short specification covering the skill’s purpose, inputs, expected outputs, and quality criteria.

2. Build and run your first skill

Create the initial skill structure and write instructions that communicate the task and its execution steps. Run it in Claude Code and compare the result with your specification.

Exercise: Build a first version and test it on a representative request.

3. Improve descriptions and activation

Test how different requests affect skill selection. Investigate descriptions that are too broad, too narrow, or unclear, and revise them based on observed behavior.

Exercise: Create a small set of requests covering intended uses and requests outside the skill’s scope.

4. Add references, scripts, and tools

Extend the workflow with supporting documents and executable operations. Explore how to communicate dependencies and permissions, and how unavailable resources can interrupt execution.

Exercise: Add a reference or scripted step, then inspect what happens when a required resource is missing.

5. Evaluate outputs and troubleshoot failures

Check results against the requirements of the task. Examine cases where an output appears successful but misses an important requirement, and distinguish problems with activation, instructions, tool execution, and evaluation.

Exercise: Identify a failure, revise the relevant part of the skill, and rerun the checks to assess the change.

6. Package and maintain the skill

Document the skill’s purpose, setup requirements, dependencies, and limitations. Discuss sharing it with colleagues and checking that updates preserve behavior that already works.

Exercise: Prepare the skill so another person can understand how to use and test it.

7. Build clinic and Q&A

Use dedicated time to refine your implementation, troubleshoot problems, and review your evaluation criteria. Discuss how to adapt the approach to other workflows and continue improving your skill after the session.

Who should attend

This masterclass is designed for developers, AI practitioners, and technical professionals who use AI assistants for recurring work and want to make their workflows easier to reuse and evaluate.

It is especially relevant if you repeatedly provide the same instructions, maintain documents an assistant needs to consult, or want to capture an established personal or team process in a skill.

Prerequisites and preparation

You should be comfortable using an AI assistant, running basic terminal commands, and editing text or code files. Previous experience building Agent Skills is not required.

To participate in the exercises, bring a laptop with Claude Code installed and working, model access configured, and a code editor such as VS Code. Basic familiarity with Python or JavaScript will help with the scripting exercises. Bring a recurring workflow you would like to turn into a skill, together with any non-sensitive examples or reference material.

What you’ll take away

• A reusable Agent Skill you have built, tested, and refined.
• A practical evaluation checklist covering activation, execution, and output quality.
• A repeatable process for diagnosing failures and assessing improvements.
• An understanding of how to package and maintain skills for continued personal or team use.

Registration

The masterclass takes place on Sunday, 6 December 2026, from 7:00 to 11:00 pm EET (UTC+2).

Places are limited to 30 participants. Book through 20 November for the $49 USD early-bird price. From 21 November, tickets are $59 USD, subject to availability.