Prompt

# Prompt

## Overview

In Batonel, a **Prompt** is the AI handoff representation of an artifact contract.

A prompt translates project context, artifact identity, and contract constraints
into a form that can be directly given to an AI coding tool or lightweight model.

The prompt is not the source of truth.
The contract is the source of truth.
The prompt is a delivery format derived from that contract.

---

## Purpose

The purpose of a prompt is to make an artifact directly implementable by an AI system.

A prompt answers questions such as:

- What should the model implement?
- What is the role of this artifact?
- What must it do?
- What must it avoid?
- What dependencies are allowed?
- What should count as complete?

Prompts reduce the need for a human to restate the same architectural context every time.

---

## Responsibilities

A prompt is responsible for:

- packaging artifact context for implementation
- restating responsibilities clearly
- restating constraints clearly
- presenting expected inputs and outputs
- defining completion expectations
- making artifact contracts usable by AI systems

A prompt is not responsible for inventing architecture.
It should reflect the contract, not replace it.

---

## Core fields or sections

Typical prompt sections include:

- artifact name
- role
- module
- responsibilities
- must not
- allowed dependencies
- forbidden dependencies
- inputs
- outputs
- completion criteria

---

## Canonical Section Order

To ensure consistency across multiple artifacts and different output modes, Batonel follows a strict canonical section order:

1.  **Identity**: Artifact Name, Role, and Module.
2.  **Scope**: Responsibilities and "Must Not" constraints.
3.  **Connectivity**: Allowed and Forbidden dependencies.
4.  **Interface**: Expected Inputs and Outputs.
5.  **Fulfillment**: Completion Criteria (Definition of Done).

This stable order makes it easier for both humans and AI models to scan and compare prompts.

Prompts may be represented as Markdown, plain text, or structured output.

---

## Example