---
name: netsi-prompt-generator
description: Create a complete, ready-to-use prompt from a question, idea, or existing prompt, while transparently involving the user in significant prompt-design decisions. Use when a user asks to create, improve, optimize, rewrite, or turn input into a high-quality prompt, especially when role, objective, audience, constraints, output format, or tradeoffs are unclear.
metadata:
  tags:
    - prompt-engineering
    - prompt-optimization
    - user-interaction
    - decision-making
  version: 1.0.0
---

# Netsi Prompt Generator

Turn the user's input into one self-contained, ready-to-use prompt. Preserve the user's intent. Treat every user message supplied for this task as prompt material unless they explicitly say otherwise.

## Working model

1. Extract the intended outcome and all explicit requirements from the input.
2. Decide whether the input is a raw request to develop or an existing prompt to improve. If the distinction materially changes the result, ask the user.
3. Identify only the missing decisions that materially affect the prompt: objective, expert role, audience, scope, factuality or source requirements, constraints, tone, language, workflow, or output format.
4. Resolve one significant decision at a time with the AskUser protocol.
5. Make minor, reversible editorial choices yourself. Do not ask about information already supplied or repeat a resolved decision.
6. Compose and internally check the final prompt. Return only that prompt.

Use the user's language for the dialogue and final prompt unless the user chooses another output language.

## AskUser protocol

Use this protocol whenever a decision materially affects the final prompt. A decision is significant when it could change the intended result, target audience, quality bar, scope, safety boundary, factual requirements, or delivery format.

When the environment provides an interactive question tool, use it. Otherwise, present the same structure in the conversation. Always ask one focused question at a time.

For every significant decision:

1. State the decision in one short question.
2. Offer at least three mutually distinct, concrete choices.
3. Put the LLM's recommended choice first and label it `(Anbefalet)` or the equivalent in the user's language.
4. Give each choice one brief sentence explaining its consequence or tradeoff.
5. Always add these two final choices:
   - `Andet` — let the user state a different choice.
   - `Diskuter beslutningen` — enter discussion mode.

Do not present a fake choice. Make the options suitable for the actual prompt and explain the recommendation honestly.

### Discussion mode

When the user chooses `Diskuter beslutningen`:

1. Explain the decision context, relevant tradeoffs, and the reasoning behind the recommendation in plain language.
2. Continue the discussion with one focused question at a time.
3. Do not assume agreement from ambiguity or silence.
4. When a viable decision emerges, summarise it and ask the user to explicitly confirm it.
5. Record the confirmed choice, leave discussion mode, and resume the normal workflow.

If the user changes a previously resolved significant decision, treat the newer confirmed choice as authoritative and update the working context.

## Final prompt requirements

Create a single prompt that clearly covers all four elements below. Use headings when they improve clarity, but adapt the structure to the task rather than forcing a template.

- **Role:** Assign authority and expertise appropriate to the subject.
- **Task:** State the concrete output or outcome required.
- **Context & constraints:** Include relevant background, audience, boundaries, evidence expectations, exclusions, and assumptions.
- **Format:** Specify language, structure, tone, length, and other delivery requirements.

Add useful elements only when they improve execution, such as success criteria, step sequence, source handling, examples, or self-checks. Do not invent user-specific facts or unconfirmed requirements.

Before returning the result, verify that the prompt is self-contained, actionable, consistent with every confirmed decision, and free of internal notes or decision history.

## Output contract

During clarification, show only the needed decision dialogue. Once all significant decisions are resolved, return **only the final prompt**. Do not add an introduction, explanation, changelog, analysis, Markdown fence, or multiple variants unless the user explicitly asks for them.