Each template is plain Markdown. Copy it, or open it in the editor, fill in the {{placeholders}} and share your version with a link. They work with ChatGPT, Claude, Gemini, Copilot and open-source models.
Structured prompt template
A general-purpose template with role, task, context, constraints, output format and an example. Start here for any model.
# Prompt Title
## Role
You are an expert assistant.
## Task
Describe exactly what you want the model to do.
## Context
Add any background the model needs.
## Constraints
- Keep the answer concise
- Ask for clarification if the request is ambiguous
## Output format
Reply in Markdown with a short summary first, then the details.
## Example
Input:
Summarize the text below in three bullet points.
Output:
- First key point
- Second key point
- Third key point
System prompt for an AI assistant
Define an assistant's identity, goals, tone and rules. Paste it into the system prompt or custom instructions of ChatGPT, Claude or Gemini.
# System Prompt: {{assistant_name}}
## Identity
You are {{assistant_name}}, a helpful assistant for {{audience}}.
## Goals
- Give accurate, practical answers
- Match the user's level of expertise
- Say so plainly when you are not sure
## Tone and style
- Friendly, direct and concise
- Use short paragraphs and bullet lists
- Avoid jargon unless the user uses it first
## Rules
- Never invent facts, figures or sources
- Ask one clarifying question when a request is ambiguous
- Decline requests that are outside {{scope}}
## Output format
Start with the answer, then add steps or details if they help.
AGENTS.md and CLAUDE.md project instructions
Project instructions for AI coding agents: commands, code style and rules. The same structure works for Cursor rules files.
# AGENTS.md
Instructions for AI coding agents working in this repository.
## Project overview
{{one or two sentences about what the project does}}
## Commands
- Install dependencies: `{{install command}}`
- Run the tests: `{{test command}}`
- Run the linter: `{{lint command}}`
## Code style
- Follow the existing patterns in the file you are editing
- Keep functions small and give things clear names
- Add or update tests with every change
## Rules
- Never commit secrets or credentials
- Ask before deleting files or changing the database schema
- Summarize what you changed and why at the end of each task
Code review prompt
Get a prioritized list of bugs, security issues and suggested fixes, returned as a Markdown table.
# Code Review Prompt
## Role
You are a senior {{language}} engineer doing a careful code review.
## Task
Review the code below and report problems in order of severity.
## Focus on
1. Bugs and edge cases
2. Security issues
3. Performance problems
4. Readability and naming
## Rules
- Quote the exact line for every issue
- Explain why it is a problem in one or two sentences
- Suggest a concrete fix as a code block
- Skip style nitpicks unless they hide a bug
## Output format
A Markdown table with the columns Severity, Line, Issue and Fix, followed by a short overall verdict.
## Code
{{paste your code here}}
Summarization prompt
Turn a long document into a one-sentence summary and five key points without losing names, numbers or dates.
# Summarize a Document
## Task
Summarize the text below for {{audience}}.
## Requirements
- Start with a one-sentence summary
- Follow with the five most important points as bullets
- Keep names, numbers and dates exactly as written
- Flag anything that looks uncertain or contradictory
## Length
No more than {{word_limit}} words.
## Text
"""
{{paste the text here}}
"""
Data extraction prompt with JSON output
Pull structured fields out of unstructured text and get valid JSON back, with null for anything that is missing.
# Extract Structured Data as JSON
## Task
Read the text below and extract the fields listed in the schema.
## Schema
```json
{
"name": "string",
"company": "string or null",
"date": "YYYY-MM-DD or null",
"amount": "number or null",
"summary": "string, at most 30 words"
}
```
## Rules
- Return valid JSON only, with no extra commentary
- Use null when a field is missing from the text
- Do not guess values that are not stated
## Text
{{paste the text here}}
Few-shot classification prompt
Teach the model your labels with a few examples, then classify new messages consistently.
# Few-Shot Classification Prompt
## Task
Classify each customer message as Praise, Complaint, Question or Other.
## Examples
Message: "The new dashboard is so much faster, thank you!"
Label: Praise
Message: "I was charged twice this month."
Label: Complaint
Message: "Can I export my data as a CSV file?"
Label: Question
## Rules
- Reply with the label only
- Pick Other when no label fits
## Message to classify
{{paste the message here}}
Image generation prompt
Describe the subject, style, composition, lighting and aspect ratio for Midjourney, Stable Diffusion and other image models.
# Image Generation Prompt
## Subject
{{main subject, for example a lighthouse on a rocky coast}}
## Style
{{art style, for example watercolor with soft pastel colors}}
## Composition
- Camera angle: {{wide shot, close-up or aerial view}}
- Lighting: {{golden hour, studio light or overcast}}
- Mood: {{calm, dramatic or playful}}
## Details to include
- {{first detail}}
- {{second detail}}
## Avoid
- Text or watermarks in the image
- Distorted hands or faces
## Aspect ratio
{{16:9, 1:1 or 9:16}}