Skills and sub-agents

A skill is a set of instructions that teaches your agent how to do a particular job well: writing in a natural voice, running market research, planning social posts. It's not software you install. Think of it as a training manual your agent reads and follows.

This page explains how skills work, how to add one, and when it helps to split a big job across sub-agents.

Skills vs software

This is the most common mix-up, so it's worth being clear:

If a skill's notes mention macOS or a desktop tool, that's usually background from wherever it was written. Your agent runs on its own Linux computer in the cloud, with no screen, so skills that are about instructions and process work well; skills that need to click around a desktop app usually don't.

How to add a skill

The easiest way is to ask your agent in chat. Our free skill library is on GitHub, with skills for writing, research, data analysis, email marketing, SEO, social media and more.

You can also hand your agent any skill folder: upload it in the Files tab and ask your agent to install it.

Test it with something small. Before a big job, ask for a tiny result using the skill, like "Using the human-writing skill, rewrite this paragraph" and check it does what you expect.

When a skill seems broken

• It needs a key or connection. Some skills rely on an app, like Notion or GitHub. Connect it on the Integrations tab first. See connect your agent to your apps.
• It needs a screen. Skills built for desktop automation won't work on a headless agent. Ask your agent to find another way to do the job.
• The agent isn't using it. Remind it: "Use the market-research skill for this." Starting a New chat after installing a skill can also help it pick it up.

Ask your agent to explain: "What does this skill need to work, and do you have all of it?"

Sub-agents: splitting a big job

Your agent can hand parts of a big task to helper agents, called sub-agents, that work at the same time. It's useful for jobs that split cleanly, like research, then a draft, then a check. It isn't worth it for small, one-step tasks.

The way to keep it from turning into chaos is one manager: your main agent assigns the work, collects the results and gives you one final answer.

1. Name one goal. "A one-page summary of the five main competitors."
2. Give each helper one job, with no overlap. Research, draft, check.
3. Ask for the same format from each. Same headings, same style.
4. Ask for one merged answer at the end, with duplicates and contradictions cleaned up.

Sub-agents each use AI, so a big split job uses more of your monthly allowance than a single agent doing a small task. See AI usage and top-ups.

Frequently asked questions

What is a skill for an AI agent?

A folder of instructions and examples that teaches your agent how to do a specific job well. It's knowledge your agent follows, not a program it runs.

Do I need to install anything on my own computer?

No. Skills live on your agent's own computer in the cloud. You just ask your agent to install them.

Where can I find skills?

The Heyron agent library on GitHub has free skills for writing, research, marketing and more. Your agent can also install skills you upload in the Files tab.

Why isn't my agent using a skill I installed?

Ask it to use the skill by name, and start a new chat if needed. If the skill needs an app connection or a key, set that up first.

When should I use sub-agents?

For big jobs that split into separate parts, like research, writing and checking. For quick tasks, one agent is faster and cheaper.