TOOLS & SKILLS
Learn Claude Skills in 10 Minutes 🛠️
Everyone overcomplicates Skills. You can build one, install it, and be using it in 10 minutes. Here's exactly how.
What Is a Claude Skill?
A Skill is a set of instructions that Claude loads automatically when it detects a matching task. Instead of re-explaining "write this email in my voice, keep it under 4 sentences, use a casual tone, and always include a clear next step" every time — you save that as a Skill and Claude applies it whenever you ask it to draft an email.
Under the hood, a Skill is just a folder containing a file called SKILL.md — a plain text file with two parts: a short description that tells Claude when to use it, and detailed instructions that tell Claude how to do the task. That's it. No code. No API. Just a text file.
Open Cowork & Launch the Skill-Creator
Open the Claude desktop app and switch to Cowork (in the left sidebar).
Make sure you're on Opus 4.6 with Extended Thinking turned on. (Click the model name at the top to switch. Extended Thinking is in the same menu.) This gives you the best skill-building results.
Type this:
Replace the bracket with something real. Examples:
- "...a skill for writing client follow-up emails in my voice"
- "...a skill for summarizing meeting notes into action items"
- "...a skill for reviewing contracts and flagging risky clauses"
- "...a skill for turning my rough notes into polished LinkedIn posts"
The Interview — Be Specific
Claude will ask you a series of questions about how you want the skill to work. This is the most important part. The more detail you give here, the better the skill performs forever.
Claude will ask things like:
- What exactly should this skill do?
- What tone or style should the output have?
- What should it always include? Always avoid?
- Can you give me an example of a good output?
When Claude is done, it generates your SKILL.md file — the text file that contains everything. You'll see a preview. Review it and tell Claude if anything feels off: "make the tone more casual" or "add a rule to always include a subject line." It'll iterate until you're happy.
Install & Test
Save the skill folder that Claude created. It'll be in your Cowork files. Download it to your computer.
Compress it into a ZIP file. Right-click the folder → Compress (Mac) or Send to → Compressed folder (Windows).
Upload it: Go to Settings → Features → Skills, click the "+" button → Upload a skill, and select your ZIP file. Toggle it on.
Test it: Open a new chat (not the same Cowork session). Type a message that should trigger your skill. Did it fire? Try different phrasings to make sure it triggers when it should.
Debug if needed: If the skill isn't firing, ask Claude:
Claude reads the skill description back to you word for word. You'll instantly see what's wrong with the trigger. Fix the description, re-upload, and test again.
Explore the Skills & Plugin Library
Now that you know how to build skills, go explore what's already been built. There are hundreds of pre-made skills and plugins ready to install.
- Plugin marketplace (in Cowork): Open the Claude desktop app → switch to Cowork → click "Customize" in the sidebar → "Browse plugins." This is Anthropic's official marketplace, launched February 2026. Install anything with one click. Each plugin often bundles 3-5 related skills.
- Connectors (your real tools): Go to Settings → Connected Apps and connect Gmail, Google Calendar, Google Drive, Slack, and other tools. These aren't skills — they're connectors that give your skills access to real data. A "morning briefing" skill is 10x better when it can actually read your calendar and inbox.
- Anthropic's official skill library: Browse Anthropic's open-source skills at github.com/anthropics/skills. Download any skill folder and upload it through Settings → Features → Skills.
- Turn any Cowork chat into a skill: Already gave Claude good instructions in a Cowork session? Click the arrow next to the chat name → "Turn into a skill." Done in 2 clicks. Every good Cowork conversation is a skill waiting to be saved.
- My 10 ready-to-use work skills: I built 10 professional skills you can copy-paste right now — status reports, meeting recaps, email drafts, and more.
What a SKILL.md File Looks Like
You don't need to write this by hand (the skill-creator does it for you), but it helps to know what you're looking at:
Instructions = how Claude does the task. Be specific. "Write a good email" is useless. "Keep it under 150 words, warm but professional, always end with a next step" is useful.
Pro Tips for Better Skills
- The description is the trigger. Spend more time on the description than the instructions. If the description is wrong, the best instructions in the world don't matter because the skill never fires.
- Always add "Do NOT use for." List everything the skill should stay away from. This one line eliminates most false triggers.
- Test immediately. Don't build 5 skills and test later. Build one, test it, fix it, move on.
- Use skills inside Projects. A skill running inside a Project gets all the Project's context for free — your files, your instructions, your preferences. Same skill, dramatically better output.
- Install everything. Having 100 skills installed doesn't slow Claude down. It only reads the full instructions when a skill matches your task. Zero cost for unused skills.
- Fix lazy output in your message, not the skill. If Claude rushes through a skill, add "Take your time. Quality over speed. Don't skip steps" to your actual message. Anthropic confirmed this works better in the prompt than in the skill file.
- Name with slash commands. Use names like /inbox, /prep, /draft, /recap. You can type the slash command to trigger the skill instantly instead of waiting for auto-detection.
- Update monthly. Your workflows change. Set a reminder to review your top 5 skills and update anything stale.
The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.