Skills vs slash commands vs CLAUDE.md
Skills, slash commands, and CLAUDE.md blur together because the content inside them can look nearly identical. The thing that separates them is dead simple: who pulls the trigger.
Same kind of content, three different triggers. A slash command fires when you type it. A skill fires when Claude matches the task. CLAUDE.md is always on.
| Who triggers it | When it loads | |
|---|---|---|
| Slash command | You — by typing /name | Exactly when you ask |
| Skill | Claude — by matching your task | Whenever it’s relevant |
| CLAUDE.md | Nobody — it’s automatic | Always, every conversation |
Pick by the trigger, not the content
The instructions inside might be word-for-word the same. The question is never what does it say — it’s when should it show up.
- Always relevant, short, every conversation needs it (your project’s basic facts)? → CLAUDE.md. (You set this up in Level 9.)
- A workflow you want to fire deliberately, on demand, by name — like
/deploy? → slash command. - Knowledge that only matters in certain situations, and you’d rather not have to remember it? → skill.
The cost of choosing wrong
Put a long, situational checklist in CLAUDE.md and it loads into every conversation — burning context on something that’s almost never relevant (remember from Lesson 10.2 how fast a conversation fills up).
Make something a skill when you really wanted to fire it precisely by name, and now you’re at the mercy of whether Claude decides it matches your task.
Matching the trigger to the need is the entire skill here.
What’s next
A skill is the one of the three you haven’t built yet — and the whole thing hinges on a single line. Let’s write one.