If you use Claude regularly for client work, there’s a good chance you’re doing the same thing over and over. Open a new chat, rebuild the same context, and recreate the same prompt you used last week. Or spend five minutes hunting through old conversations to find the setup that worked before.
This isn’t a prompting problem. It’s a workflow problem. And Claude Skills are built to solve it.
Below, we walk through exactly what a Skill is, how to build one from scratch, and how to work out which tasks in your business are actually worth creating one for.
What Is a Claude Skill?
A Claude Skill is a saved set of instructions for a specific task. You write them once. After that, Claude follows them every time you run that task — without you needing to explain yourself again.
Think about the last time you brought an associate into a piece of client work. The first thing you did was explain your approach: what the deliverable should look like, what to avoid, your particular way of working. If that task comes up regularly, you’re having that same conversation repeatedly.
A Skill is that conversation written down as a process. It’s not a chatbot, it’s not automation, and there’s no code involved. It’s a plain text file containing your instructions — nothing more.
Building One: A Live Example
To make this concrete, the video walks through building a tender assessment Skill from scratch.
The process starts inside Claude’s Customize panel. Navigate to Skills, click Create Skill, and select Create with Claude. The built-in Skill creator then guides you through defining what the Skill should do.
The key to getting a useful result is giving Claude enough context upfront, and asking it to raise any questions before it finalises the instructions. In this example, the prompt described the assessment criteria, the kind of consultancy profile to assume, and the expected output format — then asked Claude to clarify anything it needed before building.
Claude produces the Skill instructions and tests them against a realistic scenario before saving. In the tender example, it built a scoring framework across several dimensions, ran a test document through it, and returned scores, red flags, key unknowns, and a recommendation.
Once you’re satisfied, save the Skill. From that point on, running it is as simple as calling it by name in a new chat: Use the tender bid assessor to assess this opportunity. Claude recalls the instructions and applies them to whatever you pass in.
If the Skill needs adjusting over time — output format changed, new criteria added — you go back into a new chat, ask to update it, and save the revised version.
What’s Worth Building a Skill For?
Before creating a Skill for everything, it’s worth being clear about what actually makes a good candidate.
If the work changes significantly each time — different structure, different approach, different output — a Skill probably won’t help much. The same applies to tasks you do only a handful of times a year.
The pattern you’re looking for is:
- A task you do regularly
- Where the output format is broadly consistent
- Where you already have a prompt or approach you keep coming back to
Practical examples: turning client call transcripts into social content, drafting newsletters against a fixed template, running through analytics with the same structure each week. In each case — regular task, consistent output, and you already know what good looks like.
Start with one task. Build the Skill, run it a few times on real work, and adjust as needed. Don’t try to systemise everything at once.
Skills Combined With Claude Projects
One combination is worth calling out specifically.
Claude Projects let you store your business context — who you are, how your firm works, your tone of voice, the things Claude should always know. A Skill holds the instructions for a specific task.
Running a Skill inside a Project produces noticeably better outputs. The Skill isn’t just following the task instructions — it’s doing so with your full context behind it. For the tender assessment example, that means outputs calibrated to your firm rather than a generic framework.
If you’re only running Skills in standalone chats, you’re leaving a meaningful part of the value unused.
Where to Start
Pick one task you do regularly — something where you already have a rough approach or a prompt you’ve been reusing. Build a Skill around it, run it against real work a few times, and refine as you go.
Once you’ve built your first one, the pattern becomes obvious and you’ll start spotting where else it applies.
