prompt-improvement

Prompt Improvement

Turn rough intent into a clear prompt without changing what the user asked for.

Core Principle

Preserve explicit intent. Add assumptions only when labeled.

Workflow

1. Classify the problem

   • Identify whether the prompt is vague, ambiguous, overbroad, missing inputs, missing output format, missing audience, risky, or already good enough.

2. Extract intent

   • Separate explicit requirements, inferred goals, missing information, assumptions, constraints, and risks.
   • Do not invent requirements.

3. Decide whether to ask or rewrite

   • Ask 1 to 3 concise questions only when missing information changes the correct outcome.
   • Proceed with assumptions for low-risk drafting, brainstorming, research, reports, templates, and clear-enough tasks.

4. Rewrite

   • Make the task executable.
   • Add context, requirements, process, output format, quality rules, and language guidance only when useful.
   • Keep identifiers, file paths, commands, schema keys, model IDs, and API names unchanged.

5. Check fidelity

   • Confirm the rewrite preserves the user's request.
   • Remove hidden goals.
   • Mark assumptions and open questions clearly.

Default Output

## Improved Prompt

```text
[Prompt ready to use]

What Changed

Assumptions

Questions

## Reusable Prompt Skeleton

```text
Role:

Task:

Context:

Requirements:

Process:

Output Format:

Quality Rules:

Language And Tone:

Rules

• Respond in the user's language.
• Keep reusable model-facing prompt text in English when it improves portability.
• Do not make prompts longer unless the added structure improves execution.
• Do not remove explicit constraints.
• Do not convert uncertainty into fact.