Blog › AI at Work · Guides
How to set up Claude Skills
for consulting and project work
Ten free reusable skills, one set of master instructions, one Project per client. We built and deployed this exact kit for a consulting client this month. Here's the full setup, step by step, with every file free to download.
The short version
The whole setup in five lines
- A skill is a folder of instructions Claude loads when a job matches its description - the repeatable job written down once, done the same way every time.
- Upload a skill once at account level and it applies across all your conversations. No re-pasting into each Project, ever.
- One Project per client carries the per-client layer: knowledge files and instructions. Skills stay account-wide; client context stays walled off per Project.
- The master copies live in one folder on your computer. Edit there, re-upload, and every conversation runs the new version.
- The full kit is free. Ten ready-to-upload skills plus the master instructions file, download it below.
The map: where each piece lives
Three layers. Each one holds exactly one kind of thing.
Skills live here. Uploaded once.
All ten skills sit at account level and apply across all your conversations. This layer never changes between clients - it's the how.
Knowledge files and instructions live here.
Each client gets a Project holding their documents and the master instructions. Projects don't share files with each other, so nothing crosses between clients - it's the who.
The work happens here.
Open a chat inside the client's Project, ask for a skill by name, and Claude combines the account-wide method with the client-specific context.
That separation is the entire design. The method is written once and never retyped. The client context is loaded once per client and never mixed. Every chat starts with both already in place, so the first message can be the actual work.
The problem: your best prompts live in seventeen places
You've written the perfect meeting-minutes prompt. It's in a note somewhere. Also in a chat from March, a slightly different version in an email to yourself, and a third variant you rebuilt from memory last Tuesday because you couldn't find the first two. Each version produces slightly different output, so your minutes look different every week, and every new job starts with five minutes of retyping instructions the tool has already been given a dozen times.
That's not a prompting problem. It's a storage problem. The method for a repeatable job is real intellectual property, and it's been living in scrollback.
The shift: a skill is the repeatable job written down once
A skill is a folder holding one file - SKILL.md - that describes a job completely: what goes in, how the work is done, what the output looks like, and the rules that never bend. Its description tells Claude when the skill applies, so Claude brings it out on its own when the job matches - or you invoke it explicitly by naming it. Either way, the method runs identically every time, for every client, in every chat.
This is where Claude has a structural edge. ChatGPT users who build a kit like this have to copy it into every single Project by hand, then re-copy it everywhere when anything changes. Claude's skills are account-level: upload once, and they work across all your conversations automatically. Your Projects then only carry what's actually different per client. The kit becomes infrastructure instead of clipboard cargo.
The 10 skills in the kit
Four groups that follow the arc of a real engagement. Each card links to a full guide for that skill.
Run the project
Meeting Minutes
Messy transcript or rough notes to client-ready minutes you can send the same day.
How this skill works →Project Status Report
Scattered weekly updates to a one-page status report a sponsor reads in two minutes.
How this skill works →Client Proposal
Exploratory-call notes to a short outcomes-first proposal, with pricing left for you.
How this skill works →Understand the client
Discovery Interview Kit
Engagement context to a tailored interview guide, with what each question is really testing.
How this skill works →Stakeholder Persona
Discovery evidence to 1-3 evidence-grounded stakeholder personas, gaps flagged not guessed.
How this skill works →Current-State Process Map
Interview notes and documents to a written map of how the process actually runs today.
How this skill works →Make sense of it
Free download
Get the whole kit
One zip holds all ten skills, each already packaged as its own ready-to-upload zip, plus the master instructions file for your client Projects. No email gate, no signup - it's the same kit we deployed for a client this month, and the value is in how you use it, not in keeping it locked up.
- Download the kit using the button below.
- Extract it to a Skills folder somewhere sensible on your computer. That folder is your master copy from now on - every future edit starts there.
- Upload each skill zip in Claude under Customize > Skills. Once each, and you're done for good.
- Paste the master instructions into each client Project's custom instructions.
Setup: from download to first output
Six steps. The first two happen once per account, the rest once per client.
Enable code execution
In Claude, open Settings > Capabilities and turn on code execution and file creation. Skills need this switched on before the upload option appears.
Upload the 10 skills
Open Customize > Skills, press the + button, choose Create skill, then Upload a skill. Upload one zip at a time - each zip's root is the skill folder with SKILL.md inside. Ten uploads, done once, working in every conversation from then on.
Create a client Project
One Project per client. Name it after the client, not the job - the Project outlives any single engagement.
Paste the master instructions
Open the Project's custom instructions and paste in the master instructions file from the kit. This is the per-client layer that tells the skills how to behave for this client.
Add client knowledge files
Upload the engagement brief, contracts, org chart, prior reports - whatever the skills should draw on. Project knowledge stays inside this Project, so nothing bleeds between clients.
Ask for a skill by name
Start a chat inside the Project and name the job: 'Run the meeting minutes skill on this transcript.' Claude loads the skill and the client context together.
From here, every new client is steps three to six only: a Project, the instructions, their files, and a chat. The skills are already waiting. That's the payoff of the account-level layer - onboarding a client into your AI setup takes minutes, not an afternoon of pasting.
Creating your own skill
The ten skills cover the standard consulting arc, but your practice has jobs of its own - a board pack, a tender response, a workshop agenda. Any job you've done the same way three times is a skill waiting to be written. The fastest way to write one is to ask Claude to draft it. Copy this prompt and fill in the brackets:
You write Claude skills. A skill is a folder holding one SKILL.md file: YAML frontmatter with a name and a description that tells Claude when to use the skill, then the instructions. Draft a new skill for this repeatable job: [describe the job - what comes in, what should come out, who reads the output]. Structure the instructions in five sections: 1. What this does 2. What you give it 3. How to do the work 4. Output layout 5. Rules Make every instruction specific enough that the output looks the same no matter who runs it. Before drafting, ask me up to three questions if anything about the job is unclear.
When you're happy with the draft: create a new folder inside your master Skills folder, named after the skill. Save the draft into it as SKILL.md. Zip that folder, then upload the zip in Customize > Skills. Your new skill now works across all your conversations, exactly like the ten that came with the kit.
Editing a skill: the update loop
Skills improve with use. The first time a status report comes out with a section you always delete, that's not an annoyance - it's an edit waiting to be made. The loop is always the same three moves:
- Edit the master - open the skill's SKILL.md in the folder on your computer and change it there. Never anywhere else.
- Delete the old upload - in Customize > Skills, remove the current version of that skill.
- Upload the new zip - zip the edited folder and upload it. Every conversation now runs the improved version.
Replace, not add
Always delete before you re-upload. If two versions of the same skill are live at once, either one may run - and you won't know which until the output looks wrong. One master on your computer, one upload in Claude, always.
What the files actually are
Every skill in the kit is a markdown file - plain text with light markup, where a hash makes a heading and a dash makes a list item. It opens in Notepad, TextEdit, Word, or any editor you already have, so there's no special software between you and your own method. AI tools read and write markdown natively because it carries structure without any formatting baggage, which is exactly why the skills format is built on it. If you can edit a text file, you can maintain this entire kit.
Quick answers
Do I need a paid Claude plan to use skills?
No. Skills are available on every Claude plan - Free, Pro, Max, Team and Enterprise. You need code execution and file creation enabled under Settings > Capabilities, which is free to switch on. The one Free-plan limit that matters for this setup is Projects: Free accounts are capped at five, so a larger client roster eventually needs a paid plan.
What's the difference between a skill and project knowledge?
Scope. A skill is uploaded once at account level and applies across all your conversations - it carries the method, the repeatable way a job gets done. Project knowledge stays inside its Project and never leaves it - it carries the client-specific facts. The two layers are designed to combine: account-wide method, per-client context. That separation is the whole reason this kit only needs uploading once.
How is a skill different from custom instructions?
Custom instructions set standing preferences - tone, format, house rules - and they shape everything inside a Project a little. A skill is a complete procedure for one job, and Claude only loads it when the job matches the skill's description or you ask for it by name. Instructions influence everything lightly; a skill runs one thing properly. This kit uses both: skills for the ten jobs, master instructions for the standing rules.
Can my team use the same skills?
Yes, on Team and Enterprise plans - owners can provision skills across the whole organisation and enable sharing between colleagues, so everyone runs the same method. On individual plans, send a colleague the skill zips and they upload them to their own account in a few minutes. The output stays consistent either way, because the method lives in the files, not in anyone's head.
How do I update a skill after uploading it?
Edit the master file on your computer, zip the folder again, then in Customize > Skills delete the old version and upload the new zip. Replace, don't add - if two versions of the same skill are live, either one may run. The folder on your computer is always the source of truth; the uploads are just copies of it.
Does this work in ChatGPT too?
The kit does. The wiring doesn't. ChatGPT has no account-level equivalent of skills, so the same ten files have to be pasted into every Project by hand and updated in every Project when they change. We wrote a separate guide covering that exact setup - same kit, different wiring.
More from the blog
The sister post
Running ChatGPT instead? Same kit, different wiring.
The identical ten skills, set up the ChatGPT way - copied into each Project by hand, because there's no account-level layer to upload to.
On Microsoft 365?
The Copilot Cowork version of this kit
Copilot Cowork reads the same SKILL.md files. Where they go in OneDrive, what your admin has to switch on, and the fallback if Cowork isn't enabled.
Set up once. Every client after that is minutes.
Download the kit and run the six steps - or have us install and tailor it for your practice.
More from PM Project Change: the blog · free tools