A study guide to Anthropic Academy’s six-lesson Introduction to agent skills course, written in this repository as Claude/agent-skills/README.md. It is an original summary, not a transcript.1
Scope and coverage
| Lesson | Coverage in the guide |
|---|---|
| What are skills? | Article read in full; the video had no captions, so it is summarized from the article only |
| Creating your first skill | Article and video subtitles read in full |
| Configuration and multi-file skills | Article and video subtitles read in full |
| Skills vs. other Claude Code features | Article and video subtitles read in full |
| Sharing skills | Article and video subtitles read in full |
| Troubleshooting skills | Article and video subtitles read in full |
1 Upstream sources are the six course articles on anthropic.skilljar.com and their YouTube videos; this wiki cites the study guide, not the course directly.
Takeaways
- A skill is task-specific knowledge Claude discovers from its description and loads only when relevant.1 See Agent skill.
- Only names and descriptions are loaded up front; detail enters context when the task needs it.1 See Progressive disclosure.
CLAUDE.md, skills, subagents, hooks, and MCP servers each own a different job.1 See Claude Code extension mechanisms.allowed-toolsapplies least privilege to skills.1 See Least-privilege tool access.- Subagents do not inherit skills; a custom subagent lists them in its frontmatter.1 See Subagent and Subagent configuration file.
What this source changed in the wiki
It is the second Claude Code course ingested and overlaps the subagents course on tool limits, subagent configuration, and built-in agents. The two guides name the built-in agents differently; the difference is recorded on Subagent.