Quiz 2: Building and Using Skills

Test your understanding of how to build, test, and use skills following the Agent Skills Specification. Answer all 5 questions.

Question 1: Skill Directory Structure

You're building a dataset validation skill. What should your directory structure look like according to the Agent Skills Specification?

选项

• A. Just a single SKILL.md file in your project root
• B. SKILL.md in root, scripts/ for helper code, references/ for documentation, assets/ for templates
• C. Everything in a scripts/ folder with SKILL.md inside it
• D. Organize folders by agent type (claude/, codex/, opencode/)

答案解析

• A — While technically possible, this doesn't follow the specification. The spec defines a structured layout with optional scripts/, references/, and assets/ directories for organization and clarity.
• B（正确答案） — Correct! This follows the Agent Skills Specification exactly. SKILL.md is required at the root; scripts/, references/, and assets/ are optional but recommended for organization.
• C — No, SKILL.md must be at the root so agents can discover it immediately. Helper scripts go in scripts/ but the main file is at the root.
• D — No, the specification is agent-agnostic. Skills use a single format that works across all supporting agents. No agent-specific organization is needed.

Question 2: Required Frontmatter Fields

What are the only required frontmatter fields in a SKILL.md file according to the Agent Skills Specification?

选项

• A. name, description, version, and license
• B. name and description
• C. All metadata fields (name, version, author, etc.)
• D. At least 5 fields including metadata and compatibility

答案解析

• A — While useful, version and license are optional. Only two fields are required: name and description. These ensure agents can discover and understand the skill.
• B（正确答案） — Correct! The specification requires exactly these two fields. Name identifies the skill; description explains what it does and when to use it. All other fields are optional.
• C — No, many metadata fields are optional. Only name and description are required by the specification.
• D — No, the spec is minimal. Only name and description are required. Other fields like license, compatibility, and metadata are optional.

Question 3: Validating Your Skill Locally

After building your skill locally, what's the best next step to validate it?

选项

• A. Symlink it into your agent's skill directory, start the agent, and test with realistic prompts
• B. Wait until the skill is shared more broadly before you test it with an agent
• C. Run only the helper scripts and skip agent testing
• D. Read the SKILL.md manually and assume the agent will behave the same way

答案解析

• A（正确答案） — Correct! The first thing that matters is whether the agent can discover and activate the skill in a real workflow. Local install plus realistic prompts catches trigger and usability problems early.
• B — No. Validate the skill locally first so you can fix activation issues, missing files, and weak instructions before anyone else depends on it.
• C — Helper scripts matter, but they are not enough. A skill can have correct scripts and still fail because the description doesn't trigger or the instructions are unclear.
• D — No. Manual review helps, but the real test is whether the agent discovers and uses the skill correctly inside a live session.

Question 4: Helper Scripts vs. Instructions

How should you divide content between SKILL.md instructions and helper scripts?

选项

• A. Instructions contain all executable code; scripts are just for reference examples
• B. Instructions explain what to do and why; scripts contain runnable code agents can actually execute
• C. Everything goes in scripts; instructions are summaries of script functions
• D. It depends on the skill; the specification doesn't define this

答案解析

• A — No, it's the opposite. Instructions teach the strategy and steps; scripts contain actual, executable code agents run.
• B（正确答案） — Exactly right! Instructions provide strategy and guidance; scripts provide implementation. This separation makes skills teach and executable.
• C — No, instructions are critical for teaching. Without clear instructions, users don't understand when, why, or how to use the scripts.
• D — The specification and best practices do define this: instructions teach, scripts implement. This pattern is standard across skills.

Question 5: Activation Debugging

When a skill doesn't activate reliably, what should you do first?

选项

• A. Add more helper scripts immediately, even if the trigger description is vague
• B. Tighten the description and retest with both obvious and vague prompts
• C. Rename the skill folder without changing the description
• D. Turn the skill back into a long one-off prompt for every conversation

答案解析

• A — No. If the agent isn't activating the skill, the first problem is usually the triggering description or the examples, not missing helper code.
• B（正确答案） — Correct! Reliable activation comes from a clear description plus realistic prompt testing. Fix the trigger before adding more complexity.
• C — No. Folder names matter for organization, but activation usually depends on the description and examples the agent matches against.
• D — No. That throws away the reuse and progressive disclosure benefits skills are meant to provide.

---

Scoring

Count your correct answers:

• 5/5: Excellent! You understand skill development and the Agent Skills Specification. You're ready to build skills.
• 4/5: Good understanding. Review the question you missed before moving forward.
• 3/5 or less: Review the building and usage lessons before proceeding.

Summary

You've completed Quiz 2! You now understand:

• How to organize skills following the Agent Skills Specification
• Which frontmatter fields are required vs. optional
• How to validate a skill locally with a real agent
• How to divide content between instructions and executable scripts
• How to debug activation when a skill doesn't fire

These are the core skills for building and maintaining quality agent skills.

Next Steps

You've completed Unit 1: Agent Skills! You can now:

• Build skills following the Agent Skills Specification
• Use skills with Claude Code, Codex, and other agents
• Validate skills in a real project
• Write clear instructions and helper scripts

Next, we move to Unit 2: Model Context Protocol (MCP), where you'll learn how to integrate external tools and APIs with your code agents.

Ready to dive into MCPs?