Skip to content

Week 5 Reflection: Skills

Purpose

This week was about on-demand knowledge. Commands are things you remember to invoke. Skills are things the agent discovers when the situation calls for them.

Use this reflection to decide whether you understand not just how to write a skill, but when a skill is the right tool.

Short Answers

Write 3–5 sentences for each question.

1. Skill vs. Command

Pick one thing you automated or wanted to automate this week.

Would it be better as a skill, a command, or an AGENTS.md note? Explain why.

Use this rule of thumb:

  • Skill: the agent should discover it automatically.
  • Command: you want to invoke it explicitly with /name.
  • AGENTS.md: every agent needs it all the time.

2. Discovery

What words or phrases caused your skill to load?

What words or phrases failed to load it?

If you changed the description, paste the before and after versions.

3. Description Quality

Look at one skill description you wrote.

Does it include:

  • Task verbs, such as write, review, debug, audit, or summarize
  • Domain nouns, such as React component, DataFrame, API endpoint, or Docker container
  • The output the skill provides, such as a checklist, template, rubric, or troubleshooting flow

If one part is missing, rewrite the description now.

4. Skill Body Quality

When the agent loaded your skill, did it use the body well?

Choose one:

  • It followed my checklist or template closely.
  • It used the skill loosely but still improved the answer.
  • It loaded the skill but did not use it meaningfully.
  • It did not load the skill.

Explain what you would change next.

5. Judgment

Name one thing you should not turn into a skill.

Why not?

Good answers often mention one-off tasks, private secrets, unstable instructions, or information every agent should always know.

Practical Retrospective

Complete this table for the two skills you built or debugged.

Skill Best Trigger Prompt Weak Trigger Prompt Next Improvement
Skill 1
Skill 2

Exit Ticket

Before moving to Week 6, you should be able to say yes to all five:

  • [ ] I can explain the difference between a command and a skill.
  • [ ] I can describe why the skill description is the search index.
  • [ ] I can write a description with trigger phrases and domain nouns.
  • [ ] I can debug a skill that does not load.
  • [ ] I can decide when knowledge belongs in a skill vs. AGENTS.md.

Looking Ahead

Next week, you will create custom agents.

Skills make an agent smarter when it needs specific knowledge. Custom agents change the agent's role, model, and permissions. That means the safety question gets sharper: not just "What should the agent know?" but "What should this agent be allowed to do?"