🌱 Welcome to Factnetize — honest, well-researched articles on tech, health and AI. Get the weekly →
AI

Claude Skills: turn a workflow into a reusable skill

A Claude skill stores a procedure you'd otherwise paste into every chat. Create one by describing it in a conversation, save it, and test whether Claude really uses it.

Amber instruction card feeding repeated workflow steps and document outputs, illustrating skills in Claude.

⚡ Key takeaways

  • A skill saves a way of working: you explain a procedure once, and Claude follows it whenever the task comes up.
  • You can build one by talking. Describe the procedure in a chat, check what Claude writes, then save it.
  • Free supports custom skills once one setting is on.
  • Test it twice: once on a request it should catch, once on one it should ignore.

Every time you ask Claude for a status report, you paste the same block: use the headings Completed, Blocked, Next steps, and To confirm; keep it short; never invent a date. A skill stores that block for you. Ask for “a status brief” in any new chat and Claude applies your rules on its own. In this guide you create that skill by describing it in a chat, save it, and test whether Claude really uses it.

What a skill is, and what it isn’t

A skill is a saved set of instructions for a recurring task. Claude reads the short description of each skill and loads the ones that match what you ask. Anthropic’s Skills lesson shows that you can create one through conversation; you don’t have to write files by hand.

A skill differs from a project. A project holds facts for one piece of ongoing work, such as your newsletter brief. A skill holds a method you reuse across many tasks, such as how every status report is built. Facts that change belong in your request or your project, not in a skill.

Turn skills on

This works on Free, in Claude Desktop or on claude.ai. The skills guide lists the controls.

  1. Open Settings > Capabilities.
  2. Turn on Create and edit files (older layouts call it Code execution and file creation).
  3. Open Customize > Skills and look at the skills already there.

You’ll see Anthropic’s built-in skills for files such as Word, Excel, PowerPoint, and PDF. Claude uses them when you ask for those files. On a work or school account, your admin may block custom skills.

Step 1: Describe the procedure

  1. Start a new conversation.
  2. Paste the request below and send it.
  3. Answer Claude’s follow-up questions.
Help me create a custom skill named status-brief. Use it when I ask for a project status brief from notes. Produce four headings: Completed, Blocked, Next steps, and To confirm, in concise US English. Use only the facts in my notes. Never invent dates, owners, progress, or decisions. If notes conflict, ask before drafting. Keep it instruction-only, with no scripts. Interview me if anything is unclear, then create the skill.

When Claude asks what good output looks like, give an example. A good brief keeps the blocker visible and puts unknown details under To confirm. A bad brief invents a launch date to look complete. That one example teaches Claude more than adjectives like “professional”.

Step 2: Check and save the skill

Claude finishes by creating a file with your skill in it. Read it before you save it.

  1. Open the file Claude created in the chat.
  2. Check the description: it should say when to use the skill, for example “when the user asks for a status brief”.
  3. Check the instructions: the four headings and the rule about invented facts must be there.
  4. Click Save skill on the file.
  5. Open Customize > Skills and check that status-brief is listed and switched on.

No Save skill button? Download the file, then in Customize > Skills click Add and Upload skill. Attaching the file to a chat does not install it.

Step 3: Test it in fresh chats

Test in new chats, because the chat where you built the skill already knows your rules.

Project: neighborhood website refresh.
Completed: draft homepage text.
Blocked: approved photographs are missing.
Next step: review the text.
Launch date and owner: not confirmed.
  1. Start a new chat.
  2. Paste the notes and send: “Turn these notes into a project status brief.”
  3. Check for the four headings, the missing photographs, and an unconfirmed launch date and owner.
  4. Look in Claude’s visible reasoning for the name status-brief.
  5. Start another new chat and ask: “How do I use a semicolon?”
  6. Check that the answer is a normal explanation, not a status brief.

The first test shows the skill triggers on its own, without you naming it. The last test shows it stays out of unrelated work. The custom skills guide recommends checking Claude’s reasoning to confirm the skill loaded.

Step 4: Improve the method

If the brief invents an owner, fix the skill, not just this one answer. Ask Claude to update the skill so unknown owners always go under To confirm, save it again, and repeat both tests. Switch the skill off in Customize > Skills if you don’t want it in later work.

From my own use

My first skill was a travel planner that remembers my travel preferences and builds a plan for each trip. It is a first version, and every trip shows me something to tighten. That is exactly this step, repeated.

Frequently asked questions

Do I have to build a ZIP file by hand?

No. Talking it through is the easiest route. A hand-made skill folder is useful later, if you want to version or share skills.

Can I trust a skill someone else made?

Read every file first. Skills can include code that runs, so install only skills you understand.

Next: Claude connectors, which let Claude read from tools such as Google Drive. This is part 4 of Claude beyond the basics; the beginner series covers everyday prompting.

John Lock
Written by

John Lock

The Tested Weekly

One email a week with the guides we tested, sources linked. No spam.

Leave a comment

Your email address will not be published. Required fields are marked *

Weekly · No spam

The Tested Weekly

One email a week with the guides we tested, sources linked.