The Busy Person's Intro to Claude Skills
Claude has a feature that might be bigger than MCP. It's called Skills. Here's what they do and how to build one in 5 minutes.
Claude has a feature that might be bigger than MCP.
But 90% of users don't know it exists.
It's called Skills.
What Are Skills?
Skills are instruction files that teach Claude how you work.
Your code style. Your brand voice. Your processes.
Write them once. Claude loads them automatically. Forever.
Think: "custom onboarding docs for AI."
The Problem Skills Solve
Without Skills:
Chat 1: "Write in active voice, no jargon..."
Chat 2: "Write in active voice, no jargon..."
Chat 3: "Write in active voice, no jargon..."
Chat 4: "Write in active voice, no jargon..."
You repeat yourself. Claude forgets. Cycle continues.
With Skills: You just say "write the email."
┌─────────────────────────────────────────────────────────────┐
│ │
│ WITHOUT SKILLS │
│ You ────► Claude ────► Generic Output │
│ │ │
│ └── (context lost after every chat) │
│ │
├─────────────────────────────────────────────────────────────┤
│ │
│ WITH SKILLS │
│ You ────► Claude + Skills ────► Your Output │
│ (your voice, your rules, your tools) │
│ │
└─────────────────────────────────────────────────────────────┘
Build a Skill in 5 Minutes
Building a Skill takes 5 minutes:
- Create folder:
my-skill/ - Add file:
SKILL.md - Write two fields:
name: what-it-doesdescription: when to use it
- Add your instructions below
Done. Claude detects it automatically.
Where to Install
~/.claude/skills/ → Personal (available in all your projects)
.claude/skills/ → Project-specific (share via git with your team)
Drop your skill folder in either location. No commands needed.
┌──────────────────────────────────────────────────────────┐
│ │
│ your-skill-name/ │
│ │ │
│ ├── SKILL.md ◄── Required │
│ │ ┌────────────────────────────────────────────┐ │
│ │ │ --- │ │
│ │ │ name: doing-the-thing │ │
│ │ │ description: When Claude should use this │ │
│ │ │ --- │ │
│ │ │ │ │
│ │ │ # Instructions │ │
│ │ │ Step-by-step guide for Claude... │ │
│ │ └────────────────────────────────────────────┘ │
│ │ │
│ ├── scripts/ ◄── Optional │
│ ├── references/ ◄── Optional │
│ └── assets/ ◄── Optional │
│ │
└──────────────────────────────────────────────────────────┘
The Brilliant Part: Progressive Loading
Skills don't bloat your context:
- Name + description: ~50 tokens (always loaded)
- Full instructions: Only when triggered
- Reference files: Only when needed
You can bundle entire codebases. Claude only reads what's relevant.
┌──────────────────────────────────────────────────────────┐
│ │
│ LEVEL 1: ALWAYS LOADED │
│ name + description only (~50-100 tokens) │
│ │ │
│ ▼ │
│ Triggered? │
│ ╱ ╲ │
│ Yes No ──► (stays minimal) │
│ │ │
│ ▼ │
│ LEVEL 2: LOADED WHEN RELEVANT │
│ Full SKILL.md instructions │
│ │ │
│ ▼ │
│ Needed? │
│ ╱ ╲ │
│ Yes No ──► (stays lean) │
│ │ │
│ ▼ │
│ LEVEL 3: LOADED ON DEMAND │
│ scripts/ references/ assets/ │
│ │
└──────────────────────────────────────────────────────────┘
This is why skills beat mega-prompts. They stay lean until needed.
Real Example: PR Review Skill
---
name: reviewing-pull-requests
description: Use when reviewing code or PRs
---
# Review Process
1. Security check first
2. Then code quality
3. End with positive feedback
## Security Check
- SQL injection vulnerabilities
- Hardcoded secrets or API keys
- Unvalidated user input
## Code Quality
- Missing error handling
- Unclear variable names
- Functions doing too many things
## Output Format
Start with 1-line summary, then:
- Critical Issues (blocking)
- Suggestions (improvements)
- What's Good (positive feedback)
47 lines. Saves hours weekly.
Why Skills Might Be Bigger Than MCP
Here's the key insight:
MCP connects Claude to data. Skills teach Claude what to DO with that data.
MCP without Skills = powerful but generic. Skills = Claude that works like your best employee.
They compound together.
┌──────────────────────────────────────────────────────────┐
│ │
│ MCP ALONE │
│ Claude ──► Data Access ──► Generic Processing │
│ │
├──────────────────────────────────────────────────────────┤
│ │
│ MCP + SKILLS │
│ Claude ──► Data Access ──► YOUR Processing │
│ (MCP) (Skills) │
│ │
│ Data + Knowledge = Leverage │
│ │
└──────────────────────────────────────────────────────────┘
Starter Templates
Copy these and customize:
Brand Voice Skill
---
name: writing-in-brand-voice
description: Use when writing emails, docs, social posts, or any external content.
---
# Brand Voice Guidelines
## Tone
- Confident, not arrogant
- Helpful, not condescending
- Direct, not blunt
## Rules
1. Active voice always ("We built" not "It was built")
2. Sentences under 20 words
3. No jargon: "use" not "utilize", "help" not "empower"
4. Address reader as "you", never "users" or "customers"
5. End with clear next step
## Forbidden Words
leverage, synergy, revolutionary, best-in-class, empower
Commit Message Skill
---
name: writing-commit-messages
description: Use when creating git commits or writing commit messages.
---
# Commit Message Format
<type>(<scope>): <subject>
## Types
- feat: New feature
- fix: Bug fix
- docs: Documentation
- refactor: Code restructure
- test: Adding tests
- chore: Maintenance
## Rules
- Subject in imperative mood: "add" not "added"
- No period at end
- Under 50 characters
- Lowercase
## Examples
feat(auth): add password reset flow
fix(api): handle null response from payment provider
Meeting Notes Skill
---
name: formatting-meeting-notes
description: Use when summarizing meetings, calls, or discussions.
---
# Meeting Notes Format
## [Meeting Title] - [Date]
### Attendees
- [list of participants]
### Key Decisions
1. [decision and rationale]
### Action Items
| Task | Owner | Due Date |
|------|-------|----------|
| [task] | [name] | [date] |
### Next Meeting
[date/time if scheduled]
The Compound Effect
Skills compound over time:
┌──────────────────────────────────────────────────────────┐
│ │
│ WEEK 1 Brand voice skill saves 10 min/day │
│ WEEK 2 Code review skill saves 15 min/PR │
│ WEEK 3 Meeting notes skill saves 5 min/mtg │
│ WEEK 4 Commit message skill saves 2 min/commit │
│ │
│ ───────────────────────────────────────────────────── │
│ │
│ MONTH 2 You're saving 4+ hours/week │
│ MONTH 6 You've saved 100+ hours │
│ │
│ Every skill makes tomorrow more productive than today. │
│ │
└──────────────────────────────────────────────────────────┘
The 30-Second Alternative
Building skills manually works. But there's a faster path.
┌──────────────────────────────────────────────────────────┐
│ │
│ MANUAL APPROACH │
│ Think ──► Write ──► Debug ──► Structure ──► Install │
│ 5-15 minutes │
│ │
├──────────────────────────────────────────────────────────┤
│ │
│ SKILLCREATOR │
│ Describe ─────────────────────────────────► Download │
│ 30 seconds │
│ │
└──────────────────────────────────────────────────────────┘
Just describe what you want:
"Create a PR review skill that checks for SQL injection, missing error handling, and hardcoded secrets. Group output by severity. Include positive feedback at the end."
You get a complete skill package. Ready to install.
Join the waitlist for early access
TL;DR
- SKILL.md file in a folder
- Name + description = trigger
- Instructions = what Claude does
- Progressive loading = no bloat
- Works across sessions, forever
Your workflows, encoded once.
What's Next
You have two paths:
Path A: Build Manually
- Create a folder in
~/.claude/skills/ - Write your SKILL.md
- Claude detects it automatically
- Test and iterate
- Time: 5-15 minutes
Path B: Generate with SkillCreator
- Describe what you want
- Download the skill
- Drop it in
~/.claude/skills/ - Time: 30 seconds
Either way, you stop repeating yourself today.
Official templates: github.com/anthropics/skills
What process would you turn into a Skill?
Join the waitlist for early access
FAQ
How many skills can I have? No limit. Most power users have 5-15 active skills.
Do skills work with Claude Pro and Claude Code? Yes. Skills work across Claude interfaces that support them.
Can I share skills with my team? Yes. Skills are just folders. Share them via Git, Dropbox, or any file sharing tool.
What if two skills conflict? Claude uses the most relevant one based on your request. You can also specify which skill to use explicitly.
Can skills include Python scripts?
Yes. Add a scripts/ folder for deterministic tasks. Claude will run them when needed.