Most companies already have useful AI methods. They live in private chats, personal prompt habits, local notes, call prep routines, support triage shortcuts, account research patterns, and weekly reporting rituals. The issue is not that employees lack ideas. The issue is that one person's method never becomes something the whole team can run.
A company skill is the unit that fixes that. It is not a chatbot, a wiki page, or a static prompt library. It is a small, working piece of know-how that a person chose to publish, a team lead approved, and the whole company can now run in Claude.
Canonical definition: a company skill is a small, approved, versioned piece of know-how: the method in plain language, the files it needs, examples, tests, and a named owner, in the open format Claude reads natively. It does not describe the work; it does it.
What belongs inside a company skill
The method is only the start. A skill carries everything another person needs to run the work and everything the owner needs to keep it good.
| Part | What it is | Why it matters |
|---|---|---|
| Method | The way of working, in plain language: what to do, in what order, with which judgment calls. | It does not describe the work; it does it. |
| Reference files | The files the method needs: templates, definitions, policy excerpts, approved language. | Keeps output specific to your business instead of generic AI prose. |
| Examples | Real inputs and the outputs the owner considers good. | Shows Claude, and new teammates, what good looks like. |
| Tests | Checks the skill must pass before and after any change. | Protects quality when the method or the business changes. |
| Owner | A named person who approves changes and answers questions. | Every shared skill has a human accountable for it. |
| Version | Version history with a one line changelog. | You always know what changed, when, and why. |
The knacks loop
Publish, Approve, Ship, Use, Improve. The loop is lightweight but explicit, and every step is a deliberate human decision.
Describe the repeated work in plain English. knacks drafts the skill: method, reference files, examples, and tests. Publishing is deliberate. There is no ambient capture; nothing leaves your chats unless you decide it should.
The team lead approves. The domain expert, not IT, decides whether the method is right, the examples are good, and the skill is ready to share.
The approved skill lands in the whole team's Claude, nothing to install: one command for engineers, zero for everyone else. Its home is the company's skill library: plain markdown in a GitHub repository your company owns.
Skills run in Claude: web, desktop, and Claude Code. Everyone works from the approved method, on the latest version, every time.
The owner updates the skill once; the whole team is on the latest version immediately, with a one line changelog. Current by default.
Concrete examples
The best first skills are methods people already repeat. These are the candidates many teams publish first.
Turns CRM history, usage, calls, support tickets, and contract context into a renewal risk brief every CSM can run.
Support escalation skillTurns a messy ticket thread into a clean escalation summary, customer-safe response, owner, and next action.
Enterprise account research skillPackages account research, buying triggers, ICP fit, likely objections, and relevant talk tracks for revenue teams.
Weekly operating report skillBuilds a consistent operating narrative from KPIs, pipeline, support, product, project, and owner updates.
What a company skill is not
| It is not | Why not | What the skill adds |
|---|---|---|
| A prompt library | A prompt library stores text with no owner, no versions, and no way to keep it current. | Method, reference files, examples, tests, a named owner, and a version history that keeps everyone on the latest version. |
| A chatbot | A chatbot is an interface. It does not carry your team's methods. | An approved skill that runs inside Claude, the tool your team already uses. |
| A wiki page | A wiki describes work. Nobody runs a wiki page. | It does not describe the work; it does it. |
| A one-off automation | Automation may run a process, but the logic is hard to inspect and improve. | A readable method with a named owner and a version history anyone can review. |
Who needs company skills
Different leaders feel the same loss of control in different ways. Company skills give each function a practical operating layer for AI adoption.
How knacks helps
knacks turns one person's AI method into an approved skill the whole team runs in Claude. You publish the repeated work in plain English and knacks drafts the skill. The team lead reviews and approves it. The approved skill ships to the whole team's Claude, with the skill library as its home: a named owner, a version history, plain markdown in a GitHub repository your company owns. When the owner improves it, everyone is on the new version the same day. No capture and no usage tracking, ever. knacks never sees chats, screens, or who runs what.
The goal is not to review every prompt. The goal is to make sure the thing one person figured out becomes something the whole company can run, trust, and improve, and that stays current as the business changes.
Turn one private method into a company skill.
Book a walkthrough and we will pick one repeated piece of work, publish it as a draft skill, and show how it moves through approval into the whole team's Claude.
Book a walkthrough