Skills consulting for teams using AI at work

Your AI doesn't know how your company works. Yet.

Your support playbook, your pricing rules, your tone of voice, your "how we do it here". We turn them into tested AI skills that work in Claude, ChatGPT, Microsoft Copilot and whatever you switch to next.

SKILL.md
---
name: refund-requests
description: How we reply when a
  customer asks for their money back.
---

# Refund requests

- Monthly plans: refund this month.
- Annual plans: pro-rata, never in full.
- Offer the downgrade before the refund.
- Over £500? Hand to a human.
- Warm, plain English. No "unfortunately".
Tested ✓

Same request, same AI

The only difference is the skill.

you Reply to Priya. She's 5 months into an annual plan and wants a refund.

Without a skill

Hi Priya,

Unfortunately, we're sorry to hear you'd like to cancel. I've gone ahead and processed a full refund of your annual subscription, which should arrive in 5–7 business days.

As a goodwill gesture, we'd also love to offer you 3 months free if you ever decide to come back!

Best regards,
Customer Support Team

  • Promises a full refund you don't give
  • Invents a "3 months free" offer
  • Never offers the downgrade
  • Sounds like every other company

With your skill

Hi Priya,

Thanks for letting us know, and sorry it hasn't been the right fit. On annual plans we refund the unused months, which is £140 for your remaining 7.

Before I do that: if it's the price, Starter keeps all your data for £9 a month. I can switch you today instead.

Either way, just reply and it's sorted.
Jess

  • Applies your pro-rata policy correctly
  • Offers the downgrade first
  • Only promises what you actually do
  • Sounds like your team

Why this is suddenly your problem

Every AI tool now has a place to put skills. None of them write yours.

01

Your know-how lives in heads

The refund exceptions, how the founder likes proposals worded, which clients get 60-day terms. None of it is written down, so AI guesses, and it guesses generic.

02

Everyone fixes it alone

Each person corrects the same mistakes in their own chat, every day. That fix never reaches the rest of the team.

03

Pasting the handbook doesn't work

Dumping every document into a prompt quietly makes output worse. Writing skills that measurably change what AI does is a craft, and it's easy to get wrong.

How it works

Two weeks from interviews to a working library.

  1. 1

    Interview

    Short calls with the 3–5 people whose judgement you'd most like to bottle. We find the rules they apply without thinking.

  2. 2

    Gather

    We read what you already have: SOPs, templates, policies, and the emails and proposals you're proudest of. Exports are fine.

  3. 3

    Write & test

    Each skill is run against real tasks before and after. It only ships if the output measurably improves, and you see the evidence.

  4. 4

    Hand over

    Your skills, set up in the AI tools your team already uses, plus a session on keeping them current. You own every word.

Your skills. Any AI tool.

Skills are written in plain English and set up as skills, projects or custom instructions in whatever your team already pays for. There's no new app to learn and no subscription to cancel. If you switch tools next year, your skills come with you.

  • Claude
  • ChatGPT
  • Microsoft Copilot
  • Gemini
  • Claude Code
  • Cursor

Ways to work together

Start small, or have it done.

Workshop

from £1,500

Half a day, up to 8 people

  • Your team writes its first 5 skills, live
  • How to tell a good skill from a bad one
  • A template library to keep going
Ask about a workshop

Most teams start here

Skill Library Sprint

from £6,000

Two weeks, fixed scope

  • 10–20 skills written and tested
  • Before/after evidence for each one
  • Set up in the tools you already use
  • Handover session plus 30 days of fixes
Book a scoping call

Keep it sharp

from £750/mo

After a sprint

  • New skills as your processes change
  • Re-tests when AI models are upgraded
  • Monthly review of what's drifting
Ask about maintenance

Built on our own practice

"I run a portfolio of products solo. The only way that works is a library of around 50 skills, covering how I write copy, handle support, plan content and ship code, so every AI session starts knowing how I do things. SkillInvoke is that practice, pointed at your team."

Steve Farnworth, founder

Questions

Fair questions.

Do we need anyone technical?

No. Most skills are for support, sales, ops, marketing and finance, and they're written in plain English. If you also have developers using AI coding tools, we can cover them too.

Couldn't we just write these ourselves?

Yes, and the workshop exists for teams who want to. What usually happens is nobody has the time, the first attempts are too long and too vague, and nobody checks whether they helped. A sprint gets you a tested library in two weeks instead of "someday".

Which AI tools does this work with?

Claude, ChatGPT, Microsoft Copilot and Gemini for everyday work, and Claude Code, Cursor and GitHub Copilot for engineering teams. We set them up in whatever you already use.

Do you need access to our systems?

No. Exported documents, example emails and screen-shared walkthroughs all work. We're happy to sign an NDA, and nothing we write lives anywhere except your own workspace.

Isn't Microsoft / OpenAI / Anthropic going to do this?

They're building excellent places to store and share skills. They won't write yours, because only your team knows how you work. That's exactly the part we do.

Teach your AI how you work.

A 20-minute call to see whether a skill library would pay off for your team. No pitch deck.

[email protected]