10 Claude skills that turn an answer into work you can use

You ask for a cost calculation and get a nicely formatted list of numbers. You need an editable letter and receive running text. The difference sounds small, but it decides whether anyone can keep working with the result. A skill addresses exactly this gap: it gives a general-purpose AI additional instructions and resources for a particular piece of work.

The ten examples below come from Anthropic’s public skills repository. They form a task map, not a ranking. The use cases described are our own examples. The original instructions were reviewed for the selection; that does not amount to a performance test of our own.

A skill is a way of working, not a new store of knowledge

Think of a brief for a new colleague: besides the goal, she gets a template, rules for special cases and a description of the result you want. A skill can bundle information of this kind. Its existence proves neither access to all the files needed nor suitable programs or sufficient permissions. These prerequisites still belong to the actual place where the work is done.

So ask about the result first: should the deliverable keep calculating, be edited, be printed or be used in the browser? That leads to a more sensible selection than activating as many extensions as possible.

Four file tasks with a tangible end product

docx: Bullet points become a letter for volunteers with real headings and an editable task list. The first thing to do after delivery is to make a change in the actual file.
pdf: A specifically named table with page references is extracted from a document of your own. Missing values remain recognisable; an unreadable scanned area is not replaced by plausible-sounding numbers.
pptx: A project status becomes five editable slides. Each slide gets one job: starting point, development, problem, decision and next step.
xlsx: A materials list gets changeable inputs and formulas. Anyone who changes a quantity can then trace the calculation themselves.

Here, the answer in the chat is not the only thing that counts. Open the file in the intended application and change a meaningful part of it. A table range accidentally output as an image can look convincing and still prevent the further work you need.

Three skills between design and engineering

frontend-design suits an interface with a clear purpose. A usable brief names the audience, the content and the most important action. “A page for workshop registration with a recognisable date and a clear way to sign up” gives more direction than “make it modern”.

webapp-testing focuses on how a web application behaves. For a local form of your own, you can name empty, invalid and complete inputs as separate cases. A screenshot and a flow that can actually be played through answer different questions.

mcp-builder concerns building an interface for tools. Our own example would be a read-only room query with unambiguous inputs, a limited response and an understandable case with no results. A skill for developing such servers is not yet permission to use just any third-party account.

Three recurring ways of working

skill-creator helps with drafting and revising your own skills. Start with a recurring task and a few cases in which errors become visible. internal-comms suits clear team updates: done, next and blocked. Confirmed decisions and suggestions should be kept apart.

canvas-design leads to static design, such as an event poster with a concrete visual idea. Judge it at the size at which it will later be used. A strong draft does not automatically answer every question of a print production or a complete visual identity.

Start with one task rather than a collection

Choose one skill and write a small delivery contract: source data, output format, one relevant quality requirement and one behaviour that is not allowed. Example: “Create an editable task list; take over confirmed responsibilities, leave missing names open.” This lets you judge the output concretely.

Also read the relevant licence and the tools required. Publicly viewable does not automatically mean open source, particularly in the case of the document skills. A manageable selection whose results you understand is more useful day to day than a large collection without a recognisable purpose.

Sources and documentation status: 25 October 2026.