How to build a toolkit for your AI employees (an agent file, an index and a library of skills)

When I started building AI employees, I kept asking myself a question. If you can give my OpenClaw an identity and a set of skills, why aren't we doing that with the agents inside Claude or ChatGPT that we're already building?

It made perfect sense to me. You give an agent an identity, a personality and a range of skills it can pull from. So I started setting up my employees with an index of several skills instead of just one. That's how I created my COO. I made files in a folder on my computer that set out how I wanted my COO to work, and then I mapped my Claude agents to it.

That's the direction Anthropic is pointing everyone in now too - Their guidance is to start with one general agent and a library of skills, and only add more agents when a simpler setup falls short. Anthropic calls them skills. I call the set of skills an employee pulls from its toolkit.

A toolkit isn't anything fancy. It's just a folder on your computer for your AI employee, with an agent file that tells it who it is, what it connects to and which skills it has.

Building toolkits is step 4 of the Incuv8or AI Framework, the six steps I use to set up an AI team: build your AI brain, map your org, build your COO, add employees only when they pass the test, hand out the work from your task manager, and add routines where they're needed. The step before this one is building your AI COO, and the step after it is handing out the work from your task manager.

What you're building

This is the whole thing. Your org structure lives inside your AI brain. The COO sits at the top, your departments sit underneath, each employee gets its own folder, and there's one shared library of skills that every employee pulls from.

MY AI BRAIN
├── (your business, your clients, your brand)
│
└── ORG
    ├── AI ORG CHART.md
    │
    ├── COO
    │   └── agent.md                who it is + its skills index
    │
    ├── [YOUR DEPARTMENTS]          the AI Org Mapper builds these
    │                               around your business
    ├── MARKETING
    │   └── Scriptwriter
    │       └── agent.md            who it is + its skills index
    │
    └── SKILL LIBRARY
        ├── teaching-talking-head
        │   └── SKILL.md
        ├── broll-caption
        │   └── SKILL.md
        └── carousel
            └── SKILL.md

The agent file is the part that makes it an employee. Here's the shape of mine, with my business taken out.

# [EMPLOYEE NAME]

## Who I am
I'm [NAME], [MY NAME]'s [ROLE] in [DEPARTMENT]. I report to the COO.
My job is [ONE SENTENCE].
I run on [Opus / Sonnet].
I connect to: [TOOLS & ACCOUNTS].
I can read & write: [FOLDERS].
I never send, publish, pay for or delete anything without approval.

## How I work
1. Read the task.
2. Check my skills index & pick the skill that fits.
3. If nothing fits, hand it back to the COO. Don't make it up.
4. Read that skill in full & follow it.
5. Hand the work back for review.

## My skills index
| Skill | What it does | Use it when | Don't use it for |
|---|---|---|---|
| [skill] | [one line] | [the trigger] | [the near miss] |

## Learned rules
- [every correction I'm given, one line each]

The agent file tells the employee who it is, what it does, what it connects to and how it works. The index gives it the list of skills, what each one does and when to use it.

How the pieces fit together

Each skill in the library is a different process or mechanism in your business. A skill is a job description for one process, written down so AI does it your way.

Say you have one employee that handles your scriptwriting. It might write B-roll captions, talking heads, carousels and the imagery, and it pulls from a separate skill for each of those. That's how my Scriptwriter works. It's one employee, and when a task comes in it looks at what kind of post it is and picks the skill that fits.

The skills sit in one shared library rather than inside each employee, because more than one employee can use the same skill. The index in each agent file is what says which ones that employee pulls from.

This lines up with how Anthropic describes skills. A skill is a folder with a SKILL.md file inside it, and Claude only keeps each skill's name and description in mind until it needs the full thing. Your index works the same way. The employee reads the short list, and only opens the skill it's about to use. Anthropic explains it in Equipping agents for the real world with Agent Skills.

So rather than mapping out how many employees you need, you're now mapping out the repeatable processes in your business that you can train an AI agent to handle. The employees come out of that.

When a job gets its own employee

My COO is the main AI employee. Underneath it are eight departments, because I run multiple businesses. I am developing software, I have an educational and coaching platform, and I have my freelance business.

• Business Operations - legal, finance, admin and the day-to-day running of the business
• Clients - the services I deliver to my freelance clients
• Marketing - my content, social media and email
• Sales - leads through to new clients
• Product - the projects, side pieces and software I build
• Education - Tech Queen Academy
• Technology - my websites, integrations and security
• Personal Life - my concierge and my personal life

My products and my services are two separate departments. For products, I build projects, side pieces and software. Because I am a freelancer, I also have a services department. You may only need services, but I have both because of Launchpad, Tech Queen Academy and the other things I'm building on the side.

My org chart is a little more complex. However, for most of you starting out with just one stream of business - like selling digital products, coaching, or providing online services - your departments will be fewer. That is why I created the AI Org Mapper, and it's linked below.

A department is a grouping, not an employee. Inside each one, a job only gets its own employee and its own toolkit when it passes a test.

Anthropic's article When to use multi-agent systems (and when not to) says to start with one agent, and gives three reasons to add more: protecting context, running work in parallel, and specialisation. Their Claude Code docs on subagents add two more things a separate agent is good for: limiting which tools it can use, and sending work to a faster, cheaper model. Here's the test I use, built from both.

1. Does it need its own context? A big, focused area, like one product.
2. Does it need different permissions or tools? It connects to something the others shouldn't touch.
3. Does it need a different model? A cheaper one for simple work, or a stronger one for hard work.
4. Does it need to run at the same time as the others?

If the answer to all four is no, it's a skill, held by an employee you already have. Most people need a COO and three to five employees to start. I wrote up the full org chart in How to map your AI org chart before you build any AI employees, and the whole test, with examples from my own team, in Do you need another AI employee?.

What you need before you start

• The Claude desktop app on a paid plan, with the Claude Code tab.
• Your AI brain set up, because that's where the ORG folder goes. If you haven't done it, start with How to set up your AI brain in Obsidian.
• The AI Org Mapper skill: Download the AI Org Mapper skill →

You don't need to code, and you don't need to connect any tools to build a toolkit.

Step 1 - Map your org with the AI Org Mapper

If you're just getting started with AI employees, this is where it begins. Install the AI Org Mapper: open the Claude desktop app, go to Customise, then Skills, then Add Skill, and upload the SKILL.md file from the download.

Click the Claude Code tab and start a new chat. Click the + at the bottom of the chat and choose your AI brain folder. Then type this.

Hey Claude, I want to use /org-mapper to map out my AI org chart.

It asks you who you are and what you do in your business on a regular basis, one question at a time. Brain dump your answers. Voice notes are perfect for this.

What you'll see: your COO at the top, your departments underneath, the employees each one needs, the skills each employee uses, and the top five to build first. It saves it into your AI brain as AI ORG CHART.md. If you only sell services, or only sell products, tell it, and keep the departments that fit your business.

Step 2 - Put the org structure in your AI brain

Now turn the chart into folders. In the same chat, paste this.

Read my AI ORG CHART.md. Create a folder called ORG in my AI brain & move
the org chart into it. Inside ORG, create a COO folder, one folder for
each department on my chart & a folder called SKILL LIBRARY.

Don't create any employee folders or skills yet. Show me the folders
when you're done.

What you'll see: an ORG folder in your AI brain with your org chart, a COO folder, your departments and an empty skill library.

Step 3 - Build your COO's toolkit first

Your COO is the first employee you build, and it's the one that drafts everyone else with you. The full build is in The First AI Employee I'd Build as a Non-Techie Founder: An AI COO. In the same chat, paste this.

Now I want to build my COO's toolkit. Draft ORG/COO/agent.md with me:
who it is, its job, which model it runs on, what it connects to, what it
can touch, how it works & its skills index. Ask me one question at a time
& show me before you save.

What you'll see: a COO agent.md that you've read line by line before it was saved.

Step 4 - Have your COO draft your next employee

Take the first employee after the COO on your top five. Start a new chat, attach your AI brain folder, and paste this.

You're my COO. Read ORG/COO/agent.md & ORG/AI ORG CHART.md.

Before you draft anything, check [EMPLOYEE NAME] really needs to be its
own employee: its own context, different permissions or tools, a
different model, or running at the same time as the others. If it
doesn't, tell me which employee should hold the work as a skill instead.

If it does, create its folder inside its department & draft its agent.md
with the same sections as yours. The skills index lists only the skills
the org chart gives it. For each skill, write what it does, when to use
it & what it's not for. Don't write the skills yet. Show me first.

Read the index carefully. The "use it when" and "don't use it for" columns are what the employee reads to decide which skill to pick, so they need to be in your words and they need to be specific.

Step 5 - Draft one skill at a time, and test it

In the same chat, paste this.

Now draft the first skill in [EMPLOYEE NAME]'s index. Save it as SKILL.md
in its own folder inside SKILL LIBRARY.

Interview me about how I actually do this process, one question at a
time. Where I skip a step, ask me rather than guessing. Show me the
skill before you save it.

Then test it on a real piece of work.

You're [EMPLOYEE NAME]. Read your agent.md.

Here's a real job: [PASTE IT]. Tell me which skill from your index you're
using & why, then do the job.

What you'll see: the employee names the skill it picked before it does the work. If it picked the right one and the work is right, approve it and move on to the next skill. If not, tell it what was wrong. Your COO drafts each skill, and you test and approve each one.

If you want the detail on writing a good skill, it's in How to Build a Claude Skill.

What comes after the toolkits

The foundation is the part that matters most. Once your business is in your AI brain and your org structure is in place, everything else gets easy, because every employee knows who it is and where to find what it needs.

Then you build the workflow, so you can hand work to your COO without opening a chat. I use ClickUp, with a field that says which AI employee owns a task, and a small script called the dispatcher that checks ClickUp every 10 minutes and starts the right employee on it. The prompts for that are in steps 4 and 5 of the AI COO post.

Routines come last, and only where you need them, for the jobs that have to happen at the same time every day. If you're not sure whether a job needs one, I wrote up the questions I use in Claude routine or AI employee?

The catch, and the fix

The index only works if the descriptions are clear. When two skills sound alike, the employee picks the wrong one. A "captions" skill and a "B-roll caption" skill can easily get mixed up.

The fix is the "don't use it for" column. Every time an employee picks the wrong skill, tell it which one it should have used, and have it add the near miss to that skill's line in the index and a line to its Learned rules. The index gets clearer with every correction.

You picked [WRONG SKILL] for this. It should have been [RIGHT SKILL],
because [REASON]. Update your skills index so it doesn't happen again,
& add it to your Learned rules. Show me the change.

What this looks like as a business

Every business owner you support has seen the lists of agents online and doesn't know which ones they need. Mapping their repeatable processes, putting a clear org structure in their AI brain and building the toolkits with them is the part they can't do for themselves. You're paid for knowing what to build and in what order.

Start with one

If you're just getting started with AI employees, it's that simple. Download the AI Org Mapper, brain dump how you run your business, put the structure in your AI brain and build your COO's toolkit first. Then build one more employee, with one skill, and test it on a real job this week.

Download the AI Org Mapper skill →

If you haven't mapped your org chart yet, that's the foundation for all of this: start here.