Prompts

A prompt, named and versioned

A prompt is a system and user prompt pair kept as its own record — under a key, with optional labels and a schema — ready to attach to any agent and version alongside your code.

The pair

System and user, together

One record holds both halves of a prompt: the system prompt that sets the model's role, and the user prompt it works from. Give it a key and a name, and it's ready to attach to an agent.

A system prompt and a user prompt in one named record.
A unique key to reference it by.
Attach it to any agent that should follow it.

prompt · summarize-doc

system

You are a precise technical writer. Keep it plain.

user

Summarize the document in three bullet points.

Labels and schema

An optional shape

Tag a prompt with input and output labels to keep your library organized — they're plain text, yours to use as you like. And when a prompt needs a structured contract, attach a JSON schema for the input or output an agent should produce.

Input and output labels — free text, for your own organizing.
An optional JSON schema for a structured input or output.
Consumed by agents that ask for a structured contract.

schema.json

{
  "type": "object",
  "properties": {
    "summary": { "type": "string" },
    "topics":  { "type": "array" }
  }
}

In Git

Kept beside your code

Prompts live in Git, so a change to how your AI behaves is a change you can see, review, and roll back — the same way you handle everything else in the repository.

Keep exploring

A prompt, named and versioned

System and user, together

An optional shape

Kept beside your code

More on AI

AI overview

Kit

Agents

Models