Claude Code Skills & SKILL.md interview questions

28 questions from the Claude Code Skills & SKILL.md course, each with the correct answer and an explanation. Try to answer before you open the answer.

What Skills Are

1. What problem do skills mainly solve?

  1. Replacing version control
  2. Making the model larger
  3. Repeating the same procedure in every conversation without bloating the context
  4. Hosting websites
Show answer

Answer: Repeating the same procedure in every conversation without bloating the context

Skills package reusable procedures that load on demand.

Lesson: The Problem Skills Solve

2. Which two frontmatter fields does every skill need?

  1. name and description
  2. author and licence
  3. model and temperature
  4. port and host
Show answer

Answer: name and description

The name identifies the skill and the description tells Claude when to use it.

Lesson: Anatomy of a Skill: Folder, SKILL.md, Frontmatter, Body

3. What is always available to Claude about every installed skill?

  1. Only its scripts' source code
  2. Its entire body and all reference files
  3. Nothing
  4. Its name and description
Show answer

Answer: Its name and description

Only the small metadata is always loaded; the rest is loaded on demand.

Lesson: Progressive Disclosure: Why Skills Stay Cheap

4. A rule must run every single time, such as formatting a file after each edit. Which mechanism fits best?

  1. A skill description, hoping Claude remembers
  2. A hook, because it runs deterministically
  3. A longer prompt
  4. A subagent
Show answer

Answer: A hook, because it runs deterministically

Skills are guidance the model may follow; hooks are guaranteed to run.

Lesson: Skills vs CLAUDE.md, Commands, Subagents, MCP and Hooks

Writing a Good SKILL.md

5. Why prefer several small skills to one giant skill?

  1. Small skills are always faster
  2. Sharper descriptions trigger correctly and are easier to maintain
  3. Giant skills are not allowed
  4. It reduces the number of files to zero
Show answer

Answer: Sharper descriptions trigger correctly and are easier to maintain

Focused skills have focused descriptions and bodies.

Lesson: Choosing a Name and Scope

6. Which description is most likely to trigger at the right time?

  1. "Version 2 of my helper."
  2. "Helps with stuff."
  3. "A very useful skill."
  4. "Draft release notes from merged PRs. Use when asked for a changelog. Triggers on: release notes, changelog."
Show answer

Answer: "Draft release notes from merged PRs. Use when asked for a changelog. Triggers on: release notes, changelog."

It says what, when and which phrases, in concrete words.

Lesson: The Description: Your Most Important Sentence

7. Why prefer a deterministic script for a fragile step?

  1. It gives repeatable results, so the model orchestrates instead of improvising
  2. Scripts are required by the file format
  3. Models cannot read text
  4. Scripts remove the need for descriptions
Show answer

Answer: It gives repeatable results, so the model orchestrates instead of improvising

Code is predictable where free-form model output is not.

Lesson: Structuring the Body: Steps, Examples, Gotchas

Scripts, References and Templates

8. How does Claude know when to open a reference file?

  1. Files open themselves
  2. It always opens every file
  3. SKILL.md tells it which file to read and when
  4. It never opens extra files
Show answer

Answer: SKILL.md tells it which file to read and when

Clear pointers make progressive disclosure work.

Lesson: Folder Layout and Loading Rules

9. What belongs in a reference file rather than SKILL.md?

  1. The main steps every run needs
  2. The skill's name and description
  3. Long, rarely needed detail such as full API docs or schemas
  4. Nothing
Show answer

Answer: Long, rarely needed detail such as full API docs or schemas

Keep the common path in SKILL.md and push rare detail out.

Lesson: Keeping SKILL.md Short and Details in References

10. What does a skill's script give you that free-form model output does not?

  1. Deterministic, repeatable results and context-free logic
  2. Better creative writing
  3. A larger context window
  4. Automatic approval of all actions
Show answer

Answer: Deterministic, repeatable results and context-free logic

Code is predictable and its source does not consume context when run.

Lesson: Scripts: Deterministic Work Inside a Skill

11. Why keep allowed-tools narrow?

  1. Broad lists are required
  2. Narrow lists make skills run faster
  3. The field is decorative
  4. Everything listed is pre-approved, so a broad list widens what an attack or mistake can do
Show answer

Answer: Everything listed is pre-approved, so a broad list widens what an attack or mistake can do

Least privilege applies to skills like any automation.

Lesson: allowed-tools and Least Privilege

Where Skills Live and How They Run

12. Where should a skill used by the whole team on one repository live?

  1. Only in one person's home folder
  2. In the repository's project skills folder, committed to git
  3. In a chat message
  4. Nowhere; skills cannot be shared
Show answer

Answer: In the repository's project skills folder, committed to git

Project skills travel with the repo and get code review.

Lesson: Personal, Project and Plugin Skills

13. A skill never triggers automatically. What is the most useful first check?

  1. Whether the model is large enough
  2. Whether its description contains the words people actually use and says when to use it
  3. Whether the repo has a licence
  4. The colour of the terminal
Show answer

Answer: Whether its description contains the words people actually use and says when to use it

Selection depends mainly on the description matching the request.

Lesson: Automatic and Explicit Invocation

14. Why update a skill in the same PR as the procedure it describes?

  1. It makes the model faster
  2. Skills must be edited only once a year
  3. PRs cannot contain Markdown
  4. Otherwise the skill silently becomes outdated and Claude follows stale steps
Show answer

Answer: Otherwise the skill silently becomes outdated and Claude follows stale steps

Documentation that drifts from reality misleads the agent.

Lesson: Sharing, Versioning and Maintaining Skills

Testing and Improving Skills

15. Why include prompts that should NOT trigger the skill?

  1. To detect false positives that waste context or cause wrong behaviour
  2. To make the list longer
  3. Because skills must fail sometimes
  4. There is no reason
Show answer

Answer: To detect false positives that waste context or cause wrong behaviour

A skill that triggers on everything is as broken as one that triggers on nothing.

Lesson: Testing Triggering With Should and Should-Not Prompts

16. What should a command-level script test check?

  1. The colour of the output
  2. Only that the file exists
  3. The exit code and the output the agent will see, including error cases
  4. Nothing; scripts need no tests
Show answer

Answer: The exit code and the output the agent will see, including error cases

Exit codes and outputs are the script's contract with the agent.

Lesson: Testing Scripts Like Normal Code

17. What is the best source of ideas for improving a skill?

  1. Adding more text everywhere
  2. Guessing
  3. Real transcripts showing where Claude misread, skipped or wasted steps
  4. Removing the description
Show answer

Answer: Real transcripts showing where Claude misread, skipped or wasted steps

Evidence beats intuition when tuning instructions.

Lesson: Improving From Real Transcripts

Security and Trust

18. Why can scanning a skill never replace reading it?

  1. Obfuscated or novel malicious code can evade pattern matching
  2. Scanners are illegal
  3. Reading is slower than scanning
  4. Skills have no code
Show answer

Answer: Obfuscated or novel malicious code can evade pattern matching

Automated checks are a filter, not a guarantee.

Lesson: The Risks of Third-Party Skills

19. Which finding should make you refuse a skill?

  1. A template file
  2. A clear README
  3. A short description
  4. A setup step that downloads a remote script and pipes it to a shell
Show answer

Answer: A setup step that downloads a remote script and pipes it to a shell

Download-and-execute gives the source full control of your machine.

Lesson: A Review Checklist Before Installing or Merging a Skill

20. Why must a skill tell Claude to treat fetched content as data?

  1. Data is always safe
  2. Fetched pages or issues can contain injected instructions
  3. It speeds up fetching
  4. It is required by Markdown
Show answer

Answer: Fetched pages or issues can contain injected instructions

Untrusted text must not be allowed to steer the agent.

Lesson: Writing Safe Skills: Secrets, Injection and Isolation

Patterns and Recipes

21. Why must each review finding cite a file and line?

  1. It makes the report longer
  2. It forces evidence and lets unverified claims be dropped
  3. It is required by Git
  4. Findings do not need evidence
Show answer

Answer: It forces evidence and lets unverified claims be dropped

Cited evidence can be checked; vague claims cannot.

Lesson: Checklist and Review Skills

22. What does a "look at it" skill remove from the workflow?

  1. The need for a computer
  2. The need for code
  3. The need for tests
  4. Asking the human to describe what is on the screen or in the log
Show answer

Answer: Asking the human to describe what is on the screen or in the log

Claude can observe results itself and iterate without waiting for a human.

Lesson: "Look at It" Skills: Closing the Verification Loop

23. Where should a workflow skill place an approval gate?

  1. Before irreversible or external actions such as publishing or deleting
  2. After everything is already done
  3. Never
  4. Only at the very first line
Show answer

Answer: Before irreversible or external actions such as publishing or deleting

Approval is useful only before the action it guards.

Lesson: Template and Workflow Skills

24. A formatting rule must apply after every edit with no exceptions. Which is best?

  1. A longer description
  2. A sentence in a skill hoping Claude remembers
  3. A hook that runs the formatter deterministically
  4. A bigger context window
Show answer

Answer: A hook that runs the formatter deterministically

Use deterministic enforcement for rules that must always hold.

Lesson: Anti-Patterns and When Not to Use a Skill

Putting It Together

25. Why is the skill committed to the repository?

  1. To hide it from teammates
  2. Because skills only work in git
  3. So the team shares it, reviews changes like code and keeps it versioned
  4. Because it is required by Markdown
Show answer

Answer: So the team shares it, reviews changes like code and keeps it versioned

Project skills travel with the repository and get review.

Lesson: Case Study: A Release-Notes Skill for a Team

26. Your skill never triggers. You have already checked the folder name and frontmatter. What next?

  1. Rewrite the description with what, when and the exact phrases users say, then test with real prompts
  2. Make the body longer
  3. Delete the skill
  4. Rename the folder randomly
Show answer

Answer: Rewrite the description with what, when and the exact phrases users say, then test with real prompts

Selection depends mainly on the description.

Lesson: Revision: Cheat Sheet and Self-Check

27. You are asked to install a public skill whose setup says to run a downloaded script piped into bash. What do you do?

  1. Paste your tokens into it
  2. Run it immediately
  3. Pre-approve all Bash tools
  4. Refuse or inspect the script fully first; run unfamiliar code only in a sandbox without credentials
Show answer

Answer: Refuse or inspect the script fully first; run unfamiliar code only in a sandbox without credentials

Download-and-execute with your permissions is a classic compromise route.

Lesson: Revision: Cheat Sheet and Self-Check

28. Which content belongs in CLAUDE.md rather than in a skill?

  1. A rarely used 20-step deployment procedure
  2. Facts and rules needed in every session, such as the test command and conventions
  3. A large API reference
  4. A one-off migration script
Show answer

Answer: Facts and rules needed in every session, such as the test command and conventions

Always-needed context goes in memory; on-demand procedures go in skills.

Lesson: Revision: Cheat Sheet and Self-Check

← Back to the course