Prompt file imported from emil1326/AITests (.github/prompts/prompt-builder.prompt.md). Fill in {{selection}}, {{file}}, {{___}} before use. Copyright stays with the author.

Prompt Builder

Turn the request into the smallest reusable artifact that will still work in another repo.

Mission

• Decide whether the task belongs in a prompt, skill, or agent file.
• Reuse a strong public pattern when one already fits.
• Keep the artifact narrow, explicit, and easy to reuse.

Before You Start

• Read AGENTS.md, the relevant instructions, and the shared lessons file.
• Check whether the request is repeated often enough to deserve reuse.
• Look for a close public pattern before inventing a custom one.
• If the artifact is a prompt or agent, also read the local prompt and agent instruction files.

Discovery Process

1. Identity and Purpose

• Identify the intended filename.
• Define the artifact in one sentence.
• Classify it as a prompt, skill, or agent.

2. Persona Definition

• Decide what role and expertise the artifact should project.
• Keep the persona specific enough to reduce guesswork.

3. Task Specification

• Define the primary task.
• List secondary tasks only if they are truly needed.
• Name the required user input and constraints.

4. Context and Variables

• Decide whether the artifact needs {{selection}}, {{file}}, or {{___}} variables.
• Decide whether it depends on other local files or instruction files.

5. Detailed Instructions and Standards

• Write the workflow in short, direct steps.
• Include the hard limits that matter.
• Add examples when they materially reduce ambiguity.

6. Output Requirements

• Specify the file path, format, and required sections.
• State what counts as a valid result.
• Make validation steps explicit.

7. Tool and Capability Requirements

• List only the tools the artifact truly needs.
• Keep permissions tight.
• Add agent delegation only when it improves the workflow.

8. Technical Configuration

• Decide whether the artifact should run in agent, ask, or edit mode.
• Add a specific model only when the role benefits from it.

9. Quality and Validation Criteria

• Define how success will be judged.
• Call out failure modes and recovery steps.
• Keep the output contract measurable.

Builder Rules

• Decide whether the best fit is a prompt, a skill, or an agent.
• Keep the scope narrow.
• Use clear sections and concrete output rules.
• Prefer a lightly adapted public pattern over a fresh prompt from scratch.
• If the task needs repeated steps or reference assets, consider a skill.
• If the task needs isolation or a strict workflow, consider an agent.

Template Generation

Use this structure when the request should become a prompt:

---
description: "[Clear, concise outcome statement]"
agent: "[agent|ask|edit or a custom agent name]"
tools: ["[tools only when needed]"]
model: "[only if a specific model is required]"
argument-hint: "[short input hint if needed]"
---

# [Prompt Title]

## Mission
[One sentence describing the task and desired outcome.]

## Inputs
[Variables, file paths, selection context, or user-provided details.]

## Workflow
1. [Step one]
2. [Step two]
3. [Step three]

## Output
[File paths, formats, and required sections.]

## Validation
[Checks, success criteria, and failure triggers.]

Output

• Recommended file type
• Proposed filename
• Frontmatter recommendation
• Reusable section outline
• Example section or example artifact shape
• Any follow-up files that should exist
• Why the chosen structure is stronger than the generic version