Abilities & Sessions

The first page in Core Concepts. By now you’ve connected your agent and run a few commands. This page explains what’s actually happening when you do.

What is an Ability

Everything your agent does on your site is an ability — a single, defined action: read a page, update a product, list recent orders, generate an image. When you ask the agent to “find my out-of-stock products and put them on sale,” it carries that out by orchestrating several abilities in turn.

Abilities are the only way an agent can touch your site. There’s no general back-door access. This matters for two reasons: the full set of things an agent can do is a fixed, known list, and every action it does take is recorded. Nothing happens that isn’t one of these defined actions.

Abilities are grouped by area — content, WooCommerce, SEO, media, and so on. The Agent Capabilities section covers what’s available in each.

What is a Session

A session is a single working conversation between your agent and your site. The agent and site communicate over an open standard called MCP (Model Context Protocol) — the common language tools like Claude and ChatGPT use to connect to WordPress — which you set up in Connect your AI. Each new conversation you start is a new session.

This distinction matters more than it sounds: instructions and context apply within a session, so starting a fresh conversation gives the agent a clean slate.

Bootstrap: how every session begins

Before an agent can do anything on your site, it must complete one required first step, called bootstrap. This isn’t optional and it isn’t something you trigger — the agent does it automatically at the start of each session, and your site enforces it: no other action is permitted until bootstrap is done.

Bootstrap is the agent loading its briefing. In that one step it receives:

• The operational rules it must follow when working on your site.
• Your operator notes — the standing instructions you’ve written to direct how it behaves.
• A summary of any knowledge the agent has accumulated from past sessions.

Two practical consequences are worth understanding:

• Your instructions are picked up fresh each session. If you change an operator note, the next conversation you start reflects it. You don’t need to reconnect or reconfigure anything.
• A new conversation starts clean. Because bootstrap re-runs per session, the agent doesn’t carry assumptions from an earlier chat. If an agent is behaving oddly, starting a fresh conversation is often the simplest reset.

You generally won’t see bootstrap happen — on the first message of a conversation the agent completes it, then answers you — but it runs every time, and it’s the reason your rules and notes reliably take effect.

The audit log

Every change an agent makes is written to an audit log — what was done, by which user, and when. The agent can show you this log but cannot alter it. It’s the authoritative record of everything that happened, independent of what the agent tells you. You saw this in Run Your First Operation, when you asked the agent for the last few audit events.

Why it’s set up this way

Because every action is a defined ability run by a known user, and every change is logged, you can always answer “what did my agent do, and as whom?” by reading one record. And because the agent loads your instructions at the start of every session, you control how it behaves in plain language rather than through a settings panel — you write your preferences once, and the agent reads them every time. How that guidance is written and applied is covered in Ability Rules & the Handshake and The Notes System.

In summary: every action your agent takes is a defined ability; each conversation is a session that begins with bootstrap, where the agent loads your rules and notes; and every change is recorded in an audit log you can review but the agent can’t alter.