# Using agents & skills

Pick the right agent for a task, shape your own, and give them reusable skills — so the work fits how your team actually operates.

Canonical page: https://app.teachervoiceai.com/docs/guides/using-agents

An **agent** is the persona a session runs as — its instructions, the model it
uses, and the skills and tools it can reach. Every project ships with sensible
defaults, so it works out of the box. Shaping your own is how you make Machine
fit your team.

## Pick an agent for a session

When you start a [session](/docs/guides/running-a-session), the agent and model
are what determine how it behaves:

- **Agent** — choose which persona runs, from the session's settings. A
  "reviewer" agent and a "builder" agent approach the same project differently.
- **Model** — the model selector in the session header picks which Claude model
  powers the run: a faster one for quick edits, a more capable one for harder
  reasoning.

You can set these before you start, or switch mid-run.

## Browse the project's agents

Open **Customize → Agents** to see every agent available in the project. Open one
to read its full instructions — the system prompt and guidance that define how it
works. This is the quickest way to understand what each persona is for before you
run it.

## Shape your own agent

Agents are defined **as code** in your project (under the agent configuration
directory), which means they're versioned and travel with the repo. To make one
your own, you give it:

- **Instructions** — the system prompt that sets its role, voice, and how it
  should approach work.
- **Skills and tools** — which [skills](#skills) and connected
  [tools](/docs/guides/connecting-tools) it's allowed to use.

Because it's code, every change to an agent is reviewable and reversible, and the
whole team gets the same behavior.

> **Info**
> Start from the defaults and adjust. A small, clear instruction set usually
> beats a long one — give the agent the role and the guardrails, and let the
> model do the rest.

## Skills

A **skill** packages a workflow you repeat into a reusable shortcut an agent can
invoke — so instead of re-explaining a multi-step process every time, the agent
just uses the skill.

- **Browse them** under **Customize → Skills**. Open one to read what it does and
  how it's used.
- **Agents use them automatically** when a task calls for it, within the skills
  you've made available to that agent.
- **Create your own** by capturing a workflow you do often as a skill in the
  project — once it's there, every agent that's allowed to use it can.

Like agents, skills live in the project as code: versioned, shared across the
team, and compounding over time as you add more.

## Putting it together

The pattern that scales: shape a couple of **agents** around the roles your team
has, give them the **skills** for the work you repeat, connect the
**[tools](/docs/guides/connecting-tools)** they act in, and run sessions against
them. Each piece is code in your project, so your setup gets better week over
week instead of living in people's heads.

---

**Related:** [Agents concept](/docs/concepts/agents) · [Customizing your project](/docs/guides/customizing-your-project) · [Running a session](/docs/guides/running-a-session).