Claude Skills: A Practical Guide for Founders Who Actually Use AI
How to stop re-explaining yourself to Claude and start building a system that remembers.
How to stop re-explaining yourself to Claude and start building a system that remembers.
Published on 10 March 2026
Most founders I talk to have the same experience with AI. They start a conversation, explain their business, describe their audience, outline their tone of voice, give context on the project, and then ask for the thing they actually needed. Twenty minutes of setup for five minutes of output. It's one of the reasons UK SMEs are slow to adopt AI in any meaningful way.
Next session? Same thing. From scratch.
Claude Skills fix this. Not in a flashy, "AI will transform your business" way. In a quiet, practical way that saves you from repeating yourself and gets you closer to consistent, usable outputs every time you sit down to work.
I've been building and using custom Skills across my consultancy for three months now. This is what I've learned, stripped of the hype.
A Skill is a set of instructions that Claude reads before it does anything. Think of it as a brief you'd give a new team member on their first day, except Claude reads it every time, perfectly, without forgetting.
Each Skill lives in a folder. That folder can contain three things.
A SKILL.md file, which is the core instruction set. Written in plain language, not code. It tells Claude what the Skill does, when to use it, what steps to follow, and what the output should look like.
Reference files, which are the documents Claude needs to do the job properly. Your brand voice guide. Your pricing. Your client brief template. A list of phrases you never want to see in your copy. Whatever context the task requires.
Scripts or templates, which are files Claude can use or produce. Output templates, document structures, checklists.
The reference files are where the real value sits. Instead of pasting your brand guide into every conversation, Claude reads it automatically from the Skill folder. The context is loaded before you type a word.
Skills are one layer. There is also a longer play: federating scattered Markdown, journals, and operational docs so assistants can query meaning across tools, not only what sits inside one Skill folder. For that story, see what if your business answers are already in your files.
If you've used ChatGPT's Custom GPTs, you'll recognise the basic idea. But Skills go further in a few ways that matter for founders running lean teams.
Skills are folders, not just prompts. A Custom GPT is essentially a long instruction with some uploaded files. A Skill is a structured workspace. You can layer reference documents, templates, and instructions together. When I write marketing copy, Claude automatically reads my brand voice guide, my founder pain-language mapping, a style guide, and a content pillar framework before it produces a single word. Four separate documents working together, loaded without me asking.
Claude decides which Skill to use. Each Skill has a description that tells Claude when it's relevant. Ask it to write a LinkedIn post and it loads the copywriting Skill. Ask it to create a Word document and it loads the document Skill. You don't manage the routing.
They work across every Claude surface. Browser, desktop app (Cowork), and the command-line tool (Claude Code). One Skill, consistent outputs everywhere. I run the same Skills in Cowork for day-to-day work and in Claude Code for automated workflows. The instructions don't change.
Here's where the theory meets reality. I run a solo consultancy with a part-time VA. No marketing team. No in-house copywriter. Skills are how I scale the thinking without scaling the headcount.
My copywriting Skill loads four reference files before it writes anything: a brand voice guide, a founder pain-language mapping (real phrases founders use, mapped to what they actually want), content pillars, and a style guide. The output is consistently on-brand. Not perfect first time, but close enough that editing takes minutes instead of starting from scratch.
Skills reduce repetition, but they do not remove market-level sameness when everyone is prompting the same shapes. For the wider B2B "photocopy" problem, see is your marketing becoming a mere facsimile of what it could be.
The pain-language mapping is the part that makes the biggest difference. Instead of Claude guessing how a founder talks about their problems, it has a reference table. "We're competing on price and can't break the hourly billing trap" maps to positioning work. "Leads ghost after we send the quote" maps to sales narrative. The copy starts from the founder's world, not consultant vocabulary. This matters more than most people realise, because positioning that uses your customer's language consistently outperforms anything written from the inside out.
I'm dyslexic with suspected ADHD. Traditional planning systems (long documents, nested project plans, weekly reviews that take an hour) don't work for me. I built a planning Skill that runs a low-friction daily and weekly rhythm: morning check-in, evening close-out, Monday planning, Friday review. Claude leads, I talk freely, it writes the notes to my knowledge vault automatically. The whole morning check-in takes about two minutes.
I have a Skill called the Bollocks Filter. It's a 10-point checklist that catches specific writing tics: contrastive antithesis openers ("It's not about X, it's about Y"), staccato drama triplets ("Speed. Scale. Savings."), stacked rhetorical questions, empty transitions ("But here's the thing..."), and redundant superlatives. Every piece of client-facing copy runs through it before I publish. It catches the patterns that make writing sound generic and AI-generated, which is useful because some of my first drafts are AI-assisted.
At the end of every work session, I say "wrap it up" and Claude produces a structured summary: what we accomplished, what system improvements were made, and what it recommends improving. It logs the session to a changelog. This means I never lose track of decisions made or files created across sessions. When you're juggling four projects in a day, that continuity matters more than you'd expect.
This one's more advanced, but it shows where Skills can go. I've built a set of interconnected Skills that handle LinkedIn outreach: scraping search results, scoring prospects against my ideal customer profile, managing conversation state across threads, generating contextual messages, and routing all outbound communication through a central decision engine. Each Skill is a discrete workflow. They compose together because they share reference files and follow the same architectural principles.
This takes thirty seconds. Open Claude, click your profile icon, go to Settings, click Capabilities, and toggle on the Skills you want to use. Anthropic provides built-in Skills for creating Word docs, PDFs, spreadsheets, and presentations. These work well out of the box.
For custom Skills, you'll need a paid plan and access to Cowork or Claude Code. The Skill Creator (also a built-in Skill) walks you through building your own.
Pick something you do at least once a week where you're always explaining the same context. Writing your newsletter. Drafting client proposals. Producing social media posts. Creating meeting summaries.
Then write the SKILL.md. Think of it as instructions for a capable person who has never worked with you before. Be specific about what you want, what good looks like, and what to avoid. Include examples of good and bad output. The clearer the brief, the better the result.
Add your reference files. Brand guide, templates, client FAQ, style rules, whatever the task needs. These are the documents you'd normally paste into the conversation. Now they load automatically.
Test it with a real request. The first output probably won't be perfect. Tell Claude what to adjust. Most Skills take two or three rounds before they're producing what you need. That's normal. You're calibrating, not failing. If you want to see how fast this kind of iterative building can go, I wrote about building a working AI prototype in two hours using a similar approach.
I use a framework called the AI Orchestrator for thinking about how AI systems should work. Every workflow needs to answer five things: what does it know (Context), what job is it doing (Goal), how does it think (Process), what can it use (Tools), and how does it improve (Feedback).
Skills sit in the Context layer. They're how you give AI the knowledge it needs to do useful work instead of guessing. Without them, every conversation starts cold. With them, Claude already knows your brand, your audience, your standards, and your preferences before you ask for anything.
For a founder-led business without a marketing team or an operations manager or a dedicated copywriter, that's the difference between AI that creates work and AI that removes it.
They don't make Claude autonomous. You still ask for the work. Claude reads the Skill and waits. If you want AI that operates without you, that's agents and automation pipelines, a different (and more complex) layer. Skills are the instructions; automations are the trigger.
They also don't replace your judgement. A good Skill produces a strong first draft. You still need to read it, test it, and decide whether it's right. The editing is faster, but it's still editing. And as models get more capable, that judgement layer becomes more load-bearing, not less; this is the heart of the frontier model step change and why goal-oriented Skills pair naturally with stronger guardrails on anything going out the door.
And they won't compensate for unclear thinking. If you don't know your brand voice, a Skill can't invent one. If you haven't mapped your audience's pain language, Claude will guess, and it'll guess generically. The quality of your reference files determines the quality of your output.
If you're running a founder-led business with a small team, these are the ones that will save you the most time in the first month.
Brand-aware copywriter. Load your voice guide, audience pain language, and content pillars. Use it for LinkedIn posts, newsletters, blog drafts, and sales emails. This is the one that compounds fastest because you use it the most.
Meeting summariser. Feed it transcript or rough notes, get structured output: decisions made, actions with owners, and follow-ups needed. Two minutes instead of twenty.
Proposal drafter. Load your services, pricing, case studies, and a proposal template. Give it a short client brief and get a first draft that's 80% there.
Weekly planner. Especially useful if, like me, you find traditional planning systems exhausting. Build a rhythm that works for your brain. Mine takes two minutes in the morning and produces a note I can reference all day.
Content repurposer. Take one piece of content (a newsletter, a workshop recording, a voice note) and turn it into assets for multiple channels. The Skill keeps everything in your voice because it has your style guide loaded.
Skills won't transform your business overnight. They're not magic. They're a system for making AI consistently useful instead of occasionally impressive. The setup takes time. The calibration takes patience. The reference files take honest thinking about what you actually want.
But once they're running, something shifts. You stop wrestling with AI and start working with it. The outputs get closer to what you'd produce yourself. The repetitive context-setting disappears. And the time you get back isn't trivial, it compounds.
For a founder running a small team, that's worth the investment.
If you've already read our introduction to why Claude Skills matter, this guide is the next step: building real Skills that work in the day-to-day of running a business.
Chris Talintyre is the founder of Polything, a strategic marketing consultancy for founder-led businesses scaling past £2m. He builds AI-augmented systems for marketing strategy and execution.
If you're a founder thinking about how AI fits into your marketing operation without the hype, get in touch.
If you are testing this idea in your own workflow, my Claude Cowork experiment shows what "work ROI" looks like: 42 Days with Claude Cowork: £0 Revenue, Real Work Lessons.
Explore more insights and strategies to elevate your marketing approach.
Claude Skills turn organisational knowledge into reusable AI modules. For founders of £1-7m businesses, this means multiplying productivity without scaling headcount. Here is why it matters.
Brittle AI workflows break when APIs change. Learn why skills-based AI systems self-correct, cut maintenance by up to 85%, and give you a competitive edge without the 2am Slack alerts.
Most AI implementations fail because they're treated as IT deployments. Learn why AI adoption is a change management project, how to use the Human-Agent Ratio, and transform your organisation into a Frontier Firm.