Prompt Improver: Agent Skill (SKILL.md) for Claude Code, Codex and Cursor

Rewrites a rough prompt into a clear, testable one: role, task, inputs, constraints, output format and a quality check.

Prompt

Installs to .agents/skills/prompt-improver-agent-skill-skill-md-for-claude-code-codex-and-c/

mkdir -p .agents/skills/prompt-improver-agent-skill-skill-md-for-claude-code-codex-and-c && curl -sL "https://prompts-gpt.com/prompts/prompt-improver-agent-skill-skill-md-for-claude-code-codex-and-cursor/raw" -o .agents/skills/prompt-improver-agent-skill-skill-md-for-claude-code-codex-and-c/SKILL.md
SKILL.md
Preview
---
name: prompt-improver-agent-skill-skill-md-for-claude-code-codex-and-c
description: "Rewrites a rough prompt into a clear, testable one: role, task, inputs, constraints, output format and a quality check."
---

# Prompt Improver: Agent Skill (SKILL.md) for Claude Code, Codex and Cursor

Rewrites a rough prompt into a clear, testable one: role, task, inputs, constraints, output format and a quality check.

## Prompt

```
---
name: prompt-improver
description: "Rewrites a rough prompt into a clear, testable one: role, task, inputs, constraints, output format and a quality check. Use when asked to improve, fix or tighten a prompt."
---

# Prompt improver

## Steps

1. Read the prompt and name, in one line, what the person wants back.
2. List what's missing: audience, inputs, length, format, tone, a definition of done.
3. Ask at most two questions if something essential is unknowable. Otherwise assume and say what you assumed.
4. Rewrite the prompt with these parts, in order: context, task, inputs as placeholders, rules, output format, a short self-check.
5. Show the rewrite in one code block, then three bullets on what changed and why.

## Rules

- Keep the person's intent; don't add goals they didn't ask for.
- Prefer concrete numbers ("under 150 words") to adjectives ("concise").
- One task per prompt. If there are two, split them and say so.
```

## Usage Notes

Save as .claude/skills/prompt-improver/SKILL.md in one project (or ~/.claude/skills/prompt-improver/ for all of them); Codex reads .agents/skills/prompt-improver/SKILL.md. Then ask for the job in plain words and the agent loads it. Or install with: npx skills add prompts-gpt/skills --skill prompt-improver.

Example output

At a glanceStructure of the improved promptHow the rough email request was rebuilt into a testable prompt
  1. ContextCompany, customers, tenure and sender
  2. TaskWrite one email announcing a price increase
  3. InputsOld and new price, date, reason, option
  4. RulesUnder 180 words, price and date up front, no exclamation marks
  5. Output formatSubject line, email body, sign-off
  6. Self-checkVerify length, one call to action, no invented facts

Example for: the rough prompt "write me a good email to customers about our price going up, make it nice and not too long."

What you want back: a short, warm customer email announcing a price increase that keeps people from cancelling.

What's missing: who the customers are, the old and new price, the effective date, the reason for the increase, what "not too long" means, the tone boundaries, and a definition of done.

Assumed (no questions needed): a subscription product with existing paying customers, sent from the founder. Length is set to under 180 words. The facts are left as placeholders so you can fill them in.

## Context
You are writing on behalf of [COMPANY], a [PRODUCT TYPE] subscription business.
The email goes to existing paying customers who have been with us an average of
[TENURE, e.g. 14 months]. It is signed by [SENDER NAME], [SENDER ROLE].

## Task
Write one email announcing a price increase.

## Inputs
- Current price: [OLD PRICE, e.g. $12/month]
- New price: [NEW PRICE, e.g. $15/month]
- Effective date: [DATE, e.g. March 1]
- Reason (1–2 facts): [REASON, e.g. higher hosting costs, new offline mode]
- Option for customers: [OPTION, e.g. lock in current price with annual plan before Feb 15]

## Rules
- Under 180 words in the body, plus a subject line under 50 characters.
- State the new price and the date in the first two sentences.
- Give the reason plainly. No "exciting news" framing and no apologising more than once.
- Include exactly one clear action (the option above) with its deadline.
- Warm and direct tone, as if from one person. No exclamation marks.
- Do not invent features, discounts or numbers that are not in the inputs.

## Output format
Subject: <line>

<email body>

<sign-off with sender name>

## Self-check (do before answering)
- Are the price and date in the first two sentences?
- Is the body under 180 words?
- Is there exactly one call to action with a deadline?
- Does every fact appear in the inputs?
If any answer is no, revise before replying.
  • "Good," "nice" and "not too long" became testable rules. Those words now have a word limit, a tone description and a ban on exclamation marks, so you can check the output instead of judging it by feel.
  • The facts are now placeholders. The original gave the model nothing to work with, so it would have made up prices and reasons. The no-invention rule backs this up.
  • A self-check catches the most common failure. Price emails often bury the number under pleasantries. Requiring it up front and verifying that before answering keeps the email honest. It also keeps it to a single task.
Written by AI from this prompt. Your results will vary.

Save as .claude/skills/prompt-improver/SKILL.md in one project (or ~/.claude/skills/prompt-improver/ for all of them); Codex reads .agents/skills/prompt-improver/SKILL.md. Then ask for the job in plain words and the agent loads it. Or install with: npx skills add prompts-gpt/skills --skill prompt-improver.

Make it repeatable

Keep using this prompt without pasting it in every time.

Prompt details

By
Prompts-GPT editors
Published
How prompts are written and checked

Did this prompt work?

Keep exploring

More in Writing