Prompt, harness and tools

The words for what surrounds the model: the prompt it reads, the harness that runs it, and the tools it asks for.

Page 1: the big words around AI, and the model.

Token

A word-piece the model reads and writes. Context windows and usage are counted in tokens.

Context

Everything the model receives for one reply: instructions, documents, history and your message.

Context window

How much the model can "see" at once. The size of its desk, measured in tokens.

Prompt

Your message. More broadly, everything sent to the model.

System prompt

Instructions from the provider or app builder that sit above your message.

Harness

The software around the model that builds its context, runs its loop, keeps memory, asks permission and runs tools. The body.

Agent

A model in a harness that decides its own next steps and uses tools until a task is done.

Memory

Notes a harness saves and feeds back into the context later. The model itself doesn't remember.

Tool

An action the harness can take when the model asks, such as search, read a file or run code. The hands.

MCP (Model Context Protocol)

An open standard for plugging AI apps into tools and data.

Skill

A folder of instructions, scripts and resources an agent can load when needed.

Prompt injection

Text hidden in a page or file that tries to give the model orders.