💭 in your head · nothing to type
Picture the prompt you keep coming back to — code check, meeting summary, release notes.
Don't picture it as "a piece of text." Picture a falconer's hawk: it doesn't think through the attack each time — it's trained to one command and strikes the instant it's called. Your repeating prompt is already almost that hawk. Run it in your head: what exactly does it strike for you?
How many times have you given it that command by hand?
💭 in your head · nothing to type
A good reusable prompt is already almost a Skill.
Don't think of this as "saved text." Picture a locksmith who cut a key to fit one specific lock. It slides in without fumbling, on the first move — no trial and error, no friction. A Skill is that key for your task: /skill-name, and the agent opens exactly what's needed.
Notice in your head: which lock in your work is still picked by trial and error every time?
A Skill is a folder with a Markdown file SKILL.md inside: instructions that Claude Code runs when you call /name. The folder name is the command: folder meeting-summary/ → /meeting-summary.
SKILL.md structure---
name: meeting-summary
description: Summarizes a meeting from a transcript or notes
---
# Meeting Summary Skill
## Goal
Create a structured meeting recap.
## Instruction
You are a meeting assistant. Based on the provided data:
1. Pull out the main decisions (3–5 bullets)
2. Identify tasks with owners and deadlines
3. Note open questions
## Output format
### Decisions
- [list]
### Tasks
- [ ] [task] — [owner] — [deadline]
### Open questions
- [list]
~/.claude/skills/ — personal, work across all projects: ~/.claude/skills/<name>/SKILL.md.claude/skills/ — project-scoped, in git: .claude/skills/<name>/SKILL.md