Claude · Foundations
A skill is the difference between typing a prompt and pressing a button.
Chatting with an AI tool works fine once. The problem is the tenth time, when someone has to remember the exact phrasing that worked last week, or retype three paragraphs of context just to get a usable draft. A skill is what turns that one-off prompt into something your team can use without thinking about it.
Direct answer
A skill is one task, built once, run the same way every time
Not a chatbot personality. A specific input, a specific set of rules, and a specific output.
A Claude skill takes one recurring task, something your team already does by hand, and turns it into a tool with a fixed shape: a defined input (a voice note, a form, a set of notes), a set of rules for how to handle it (your tone, your policies, your format), and a defined output (a drafted email, a formatted record, a summary). Someone drops in the input and gets the same quality of result every time, instead of a different outcome depending on how well the prompt was written that day.
What's actually inside one
Four parts, every time
The trigger and input
What starts it, and what gets dropped in: a technician's voice note, a lead's intake form, a set of meeting notes.
The rules
Your tone, your policies, your formatting standards, pulled from the company knowledge base so the skill doesn't have to guess.
The output
A specific, usable thing: a drafted email, a formatted work order, a clean summary, not a wall of generic text to edit down.
The review step
A person checks it before it reaches a customer, a record, or a system. The skill drafts. It doesn't send.
What this looks like
The same task, before and after it's a skill
An illustrative example, not a real business.
"Write a follow-up email for a customer who got an estimate for a heat pump replacement, keep it friendly but not pushy, mention financing is available, our name is..." retyped, slightly differently, by whoever's turn it is to write it.
Drop in the customer's name and the quote details. Get back a follow-up drafted in the company's actual voice, referencing the actual equipment quoted, formatted the way it's always formatted. Every time.
Nothing about the underlying AI changed between the two. What changed is that the second version has the company's rules and format built in, so getting a good result doesn't depend on someone writing a good prompt from memory.
Where it fits
This is session 2 of the AI Concierge
The first working session builds the knowledge base. The second builds the first skill on top of it, and every session after adds one more.
Skills aren't built in a vacuum. They draw on the company knowledge base set up first, which is why a skill answers with your actual pricing and policies instead of a generic guess. We rank the next skill to build by time saved, build it with your team in the room, and refine it as your business changes.
Questions
FAQ
Is a skill the same thing as a custom chatbot?
No. A chatbot answers open-ended questions. A skill handles one specific, repeated task with a defined input and output, which is what makes it consistent.
Can a skill send something without anyone checking it?
Not for anything customer-facing, financial, or legal. A skill drafts; a person on your team reviews and sends. That's a standing rule, not a setting to turn off.
How many skills does a typical business end up with?
It varies, but most clients have several within the first few months, built one per session in order of time saved, starting with whatever eats the most hours today.
Next step
Let's find the task worth turning into your first skill
A short call to look at what your team retypes or redrafts most often, and whether it's a good fit for a skill.