Lesson 1 - Why Agent Skills Matter

Diagnose the high-friction workflows in your day, understand what agent skills are, and learn how the open standard lets one skill work across 30+ AI coding tools.

Duration: 1.5-2 hours

Learning Objectives

By the end of this lesson, you will be able to:

✓Identify repeated AI workflows that cost you time and produce inconsistent results.
✓Define what an agent skill is and how it differs from a paste-in prompt or checklist.
✓Explain how the open standard keeps skills portable across 30+ tools.
✓Differentiate personal, project, and enterprise skills and know when each level is appropriate.

Videos

The Repetition Tax: Why Skills Exist

Ground the lesson in real pain: capture the wasted minutes, inconsistent outputs, and lost context that accumulate when you retype the same instructions.

Duration: 8 minutes

Video coming soon

What Is an Agent Skill? (From 10,000 ft)

Define the artifact precisely: a named folder with a SKILL.md file, optional supporting files, and a clear contract the AI follows.

Duration: 9 minutes

Video coming soon

Skill Tiers: Personal, Project, and Enterprise

Explain where skills live, who they affect, and when to promote a skill from personal experimentation to team-wide standard.

Duration: 6 minutes

Video coming soon

Key Concepts

The Skill Candidate Scorecard

Before writing a skill, score the workflow you want to automate. Assign 0-5 for each dimension: | Dimension | What to Assess | 0 (Low) | 5 (High) | |-----------|---------------|---------|----------| | **Frequency** | How often do you repeat this? | Once a month | Multiple times daily | | **Variance** | How inconsistent are the results? | Always the same | Different every time | | **Risk** | What happens when a step is missed? | Minor inconvenience | Downtime or rework | | **Teachability** | Can you describe the workflow clearly? | Too complex to explain | Clear checklist exists | **Prioritize workflows scoring 12+.** These are your best skill candidates. Capture this score now — you'll compare it to your results after the capstone to see the impact.

Precise Definition

An **agent skill** is a folder containing a `SKILL.md` file (metadata + instructions) and optional supporting directories. AI tools read the metadata first, then load instructions and resources on demand. This enables: 1. **Automatic activation** — the description tells the AI when to load the skill. 2. **Consistent results** — the same instructions produce the same quality every time. 3. **Portability** — one folder works in Claude Code, Copilot, Cursor, Gemini CLI, VS Code, JetBrains, and more.

Running Example: The deploy-checklist Skill

Throughout this course, we'll build one skill together: **deploy-checklist**. **The problem:** Before shipping code, you need to verify migrations, run tests, check environment variables, and notify the team. People forget steps. Different team members check different things. **The skill:** A reusable checklist your AI assistant runs through before every deployment. **Where we're going:** - Lesson 1: Identify the problem and draft a charter - Lesson 2: Create the folder structure and SKILL.md skeleton - Lesson 3: Write the instruction body - Lesson 4: Deploy it as a project skill - Lesson 5: Add a reference template and validation script - Lesson 6: Add Claude Code auto-activation for deploy-related files - Lesson 7: Validate, cold test, and ship You'll also build your OWN skill alongside this example.

Skills vs. Other Approaches

| Approach | Reusable? | Shareable? | Cross-platform? | Includes files? | Auto-activates? | |----------|-----------|-----------|-----------------|----------------|----------------| | One-off prompt | No — copy/paste | Manual | No | No | No | | System instructions | Per-session | Not easily | No | No | Yes | | CLAUDE.md / .cursorrules | Per-project | Via git | Tool-specific | No | Yes | | **Agent Skill** | **Yes** | **Via git** | **30+ tools** | **Yes** | **Yes** | Skills are the only approach that combines reusability, portability, and bundled resources.

Skill Tiers and Ownership

| Tier | Where It Lives | Who Uses It | When to Use | |------|---------------|-------------|------------| | Personal | `~/.claude/skills/` (or equivalent) | Just you | Preferences, personal workflows, experimentation | | Project | `.claude/skills/` inside repo | Your team | Team conventions, shared checklists, code review standards | | Enterprise | Org-managed directory or registry | Everyone | Compliance checks, company-wide standards, security policies | Start personal. Promote when multiple people depend on the same behavior.

Common Mistakes & Pitfalls

❌ Thinking skills require programming knowledge

Skills are written in plain markdown — the same format as a README file. If you can write a to-do list, you can write a skill.

❌ Trying to automate everything at once

Your first skill should solve ONE workflow. A focused skill that handles one thing well is more valuable than a sprawling skill that handles ten things poorly.

❌ Assuming skills only work in one tool

Agent Skills is an open standard. A skill written for Claude Code also works in Copilot, Cursor, and 30+ other tools without modification.

❌ Confusing skills with plugins or extensions

Skills are just instruction files — no installation, no API keys, no app store. Drop a folder in the right place and it works.

❌ Skipping the scorecard and jumping straight to building

Without scoring your candidate workflow, you might automate something that doesn't save meaningful time. Five minutes on the scorecard prevents hours of wasted effort.

Exercises

Exercise 1: Workflow Baseline Audit

20 minutes

List 3-5 workflows where you repeatedly brief an AI assistant. Score each using the scorecard (frequency, variance, risk, teachability).

Expected Output:

A scored table showing your top skill candidates ranked by total score.

Success Criteria:

•Listed at least 3 repeated workflows with specific details.
•Scored each on all 4 dimensions (frequency, variance, risk, teachability).
•Identified one workflow scoring 12+ as your capstone candidate.

Exercise 2: Skill Charter Draft

25 minutes

For your top-scoring workflow, write a one-page charter: problem statement, who it helps, desired outcome, and how you'll measure success.

Expected Output:

A skill charter document you'll reference throughout the course.

Success Criteria:

•Problem statement describes the pain in concrete terms (not 'it would be nice').
•Describes the trigger scenario clearly (e.g., 'when reviewing backend PRs' or 'before each weekly report').
•Lists any templates, scripts, or references you already use for this workflow.

Exercise 3: Landscape Scan

15 minutes

Visit at least two community skill repositories (github.com/anthropics/skills, github.com/github/awesome-copilot) and find skills similar to your charter.

Expected Output:

Notes on 2-3 existing skills: what they do well, what you'd change, and ideas to borrow.

Success Criteria:

•Found and linked at least 2 existing community skills.
•Noted one idea to borrow and one pitfall to avoid.
•Summarized how these findings influence your charter.

Lesson Reflection

Take a moment to reflect on what you've learned:

1. Think about the last week of using your AI assistant. Which instructions did you repeat more than twice?
2. Score your top three repeated workflows using the scorecard. Which scored highest and why?
3. If you could teach your AI one thing it would remember forever, what would it be?
4. How would your team benefit if everyone's AI assistant followed the same conventions?

Resources & Further Reading

📖Agent Skills Open Standard→📖Claude Code Skills Documentation→📖GitHub Copilot Agent Skills→📖Claude Platform Agent Skills Overview→