Writing Your First Prompt

In Getting Started, you wrote a simple prompt with a few tags and ran it. This page walks you through all the building blocks you can use to structure your prompts.

The Building Blocks

Every prompt is made of tags that describe different parts of your instructions to the AI. Here are the main ones:

Tag	Purpose
<Prompt>	Wraps your entire prompt
<Role>	Who the AI should be
<Task>	What the AI should do
<Context>	Background information
<Constraint>	Rules the AI should follow
<Format>	How the output should look
<Section>	Custom labeled sections
<Audience>	Who the output is for
<Tone>	The emotional quality of the output

You don't need all of them in every prompt. Use the ones that make sense for your situation.

Roles

The <Role> tag tells the AI who it should be. You can write a custom role description or use a preset.

Custom Role

Write your own description directly inside the tag:

<Prompt name="helper" bare>
  <Role>You are a patient math tutor who explains concepts step by step.</Role>
  <Task>Help the user solve their math problem.</Task>
</Prompt>

This produces:

<role>
You are a patient math tutor who explains concepts step by step.
</role>
<task>
Help the user solve their math problem.
</task>

Role Presets

Instead of writing your own role description, you can use a preset. Presets are pre-built personas based on prompt engineering best practices:

<Prompt name="code-review" bare>
  <Role preset="engineer" />
  <Task>Review this code.</Task>
</Prompt>

The preset="engineer" expands to:

<role>
You are a senior Software Engineer with expertise in software development,
programming, system design. You are analytical, detail-oriented, problem-solver.
</role>
<task>
Review this code.
</task>

There are presets for many roles:

Preset	Title
assistant	Assistant
engineer	Software Engineer
writer	Writer
analyst	Business Analyst
teacher	Teacher
designer	Designer
data-scientist	Data Scientist
consultant	Consultant

See the full list at Presets.

Tasks

The <Task> tag tells the AI what to do. Write your instructions directly inside it:

<Prompt name="summary" bare>
  <Task>Summarize the following document in 3 bullet points.</Task>
</Prompt>

This produces:

<task>
Summarize the following document in 3 bullet points.
</task>

Context

Use the <Context> tag to give the AI background information it needs:

<Prompt name="review" bare>
  <Context>
    The application is a REST API built with Express.js.
    It handles user authentication and data storage.
  </Context>
  <Task>Suggest improvements to the architecture.</Task>
</Prompt>

This produces:

<context>
The application is a REST API built with Express.js.
It handles user authentication and data storage.
</context>
<task>
Suggest improvements to the architecture.
</task>

Constraints

Constraints set rules for the AI to follow. Use the type property to indicate how strict each rule is:

<Prompt name="docs" bare>
  <Constraint type="must">Explain each issue clearly.</Constraint>
  <Constraint type="should">Include code examples.</Constraint>
  <Constraint type="must-not">Do not suggest removing tests.</Constraint>
</Prompt>

This produces:

<constraint>
MUST: Explain each issue clearly.
</constraint>
<constraint>
SHOULD: Include code examples.
</constraint>
<constraint>
MUST NOT: Do not suggest removing tests.
</constraint>

Here are the available constraint types:

Type	Meaning
must	Required — the AI must follow this rule
should	Recommended — the AI should follow this if possible
must-not	Forbidden — the AI must not do this
may	Optional — the AI can do this if appropriate
should-not	Discouraged — the AI should avoid this

Format

The <Format> tag tells the AI how to structure the output:

<Prompt name="report" bare>
  <Task>Write a summary of the meeting.</Task>
  <Format type="markdown" />
</Prompt>

This produces:

<task>
Write a summary of the meeting.
</task>
<format>
Output format: markdown
</format>

The available format types are markdown, json, xml, text, code, yaml, csv, list, and table.

Sections

When you need a section that doesn't match any built-in tag, use <Section> to create your own:

<Prompt name="review" bare>
  <Section name="background">
    This project uses TypeScript and React.
  </Section>
  <Task>Review the code.</Task>
</Prompt>

This produces:

<background>
This project uses TypeScript and React.
</background>
<task>
Review the code.
</task>

Audience and Tone

Tell the AI who the output is for and how it should sound.

You can describe your audience as free-form text, or use the level and type properties for built-in guidance:

<Prompt name="tutorial" bare>
  <Audience level="beginner" type="general" />
  <Tone type="friendly" formality="informal" />
  <Task>Explain how REST APIs work.</Task>
</Prompt>

The <Audience> tag produces:

<audience>
Target audience: beginner general users

Use simple language, avoid jargon, and provide analogies where helpful.
</audience>

The <Tone> tag produces:

<tone>
Tone: friendly
Be warm, approachable, and supportive.
Voice characteristics: formality: informal
</tone>

Smart Defaults

When you write a <Prompt> without certain sections, pupt fills in sensible defaults. For example, a minimal prompt:

<Prompt name="test" noRole>
  <Task>Summarize this document.</Task>
</Prompt>

Pupt fills in the format and constraints for you:

<task>
Summarize this document.
</task>
<format>
Output format: markdown
</format>
<constraints>
- Keep responses concise and focused
- Be accurate and factual
- Acknowledge uncertainty when unsure
</constraints>

If you don't include a <Role>, pupt adds one using the "assistant" preset. It also adds a default <Format> section and a <Constraints> section with sensible default rules. You only get these defaults when you leave those sections out -- if you provide your own, pupt uses yours instead.

Controlling Defaults

You can turn off individual defaults:

<!-- Skip the default role -->
<Prompt name="test" noRole>
  <Task>Help the user.</Task>
</Prompt>

<!-- Skip the default format -->
<Prompt name="test" noFormat>
  <Task>Help the user.</Task>
</Prompt>

<!-- Skip the default constraints -->
<Prompt name="test" noConstraints>
  <Task>Help the user.</Task>
</Prompt>

Or turn off everything at once with bare:

<Prompt name="simple" bare>
  <Role>You are a helpful assistant.</Role>
  <Task>Answer the user's question clearly and concisely.</Task>
</Prompt>

This produces only what you write:

<role>
You are a helpful assistant.
</role>
<task>
Answer the user's question clearly and concisely.
</task>

Examples and Reasoning

Providing Examples

You can show the AI exactly what you expect with input/output examples:

<Prompt name="test" bare>
  <Task>Answer factual questions briefly.</Task>
  <Examples>
    <Example>
      <ExampleInput>What is 2+2?</ExampleInput>
      <ExampleOutput>4</ExampleOutput>
    </Example>
    <Example>
      <ExampleInput>What is the capital of France?</ExampleInput>
      <ExampleOutput>Paris</ExampleOutput>
    </Example>
  </Examples>
</Prompt>

This produces:

<task>
Answer factual questions briefly.
</task>
<examples>
<example>
<input>
What is 2+2?
</input><output>
4
</output>
</example>
<example>
<input>
What is the capital of France?
</input><output>
Paris
</output>
</example>
</examples>

Step-by-Step Reasoning

Guide the AI through a problem one step at a time:

<Prompt name="test" bare>
  <Steps>
    <Step>Understand the requirements</Step>
    <Step>Break the problem into parts</Step>
    <Step>Solve each part</Step>
    <Step>Verify the solution</Step>
  </Steps>
</Prompt>

This produces:

Think through this step by step.

<steps>
1. Understand the requirements
2. Break the problem into parts
3. Solve each part
4. Verify the solution
</steps>

<Steps> also has presets for common reasoning patterns:

<!-- Structured analysis: Understand → Analyze → Conclude -->
<Steps preset="analysis" />

<!-- Problem solving: Define → Explore → Solve → Verify -->
<Steps preset="problem-solving" />

<!-- Debugging: Reproduce → Isolate → Fix → Verify -->
<Steps preset="debugging" />

Chain of Thought

For a simpler approach, <ChainOfThought> encourages the AI to show its reasoning:

<Prompt name="test" bare>
  <ChainOfThought>
    Think about the problem carefully before answering.
  </ChainOfThought>
</Prompt>

This produces:

<reasoning>
Think about the problem carefully before answering.
Show your reasoning process.
</reasoning>

Edge Cases and Fallbacks

You can tell the AI what to do in unusual situations:

<Prompt name="test" bare>
  <EdgeCases>
    <When condition="the user provides no code"
          then="ask them to paste the code they want reviewed" />
    <When condition="the code is in an unfamiliar language"
          then="let the user know and suggest a general review" />
  </EdgeCases>
</Prompt>

This produces:

<edge-cases>
<when>
When the user provides no code: ask them to paste the code they want reviewed
</when>
<when>
When the code is in an unfamiliar language: let the user know and suggest a general review
</when>
</edge-cases>

Both <EdgeCases> and <Fallbacks> have presets for common patterns:

<!-- Handles: missing data, out-of-scope requests, ambiguity -->
<EdgeCases preset="standard" />

<!-- Handles: unable to complete, missing info, errors -->
<Fallbacks preset="standard" />

Putting It All Together

Here's a complete prompt that brings together many of the tags you've learned:

<Prompt name="writing-assistant">
  <Role preset="writer" />
  <Task>Help the user improve their writing.</Task>
  <Context>
    The user is working on a blog post about technology trends.
  </Context>
  <Constraint type="must">Preserve the author's voice and style.</Constraint>
  <Constraint type="should">Suggest specific improvements, not vague advice.</Constraint>
  <Constraint type="must-not">Rewrite entire paragraphs without being asked.</Constraint>
  <Audience level="intermediate" type="general" />
  <Tone type="friendly" formality="semi-formal" />
  <Format type="markdown" />
</Prompt>

This produces:

<role>
You are an expert Writer with expertise in content creation, storytelling,
communication. You are creative, articulate, thoughtful.
</role>
<task>
Help the user improve their writing.
</task>
<context>
The user is working on a blog post about technology trends.
</context>
<constraint>
MUST: Preserve the author's voice and style.
</constraint>
<constraint>
SHOULD: Suggest specific improvements, not vague advice.
</constraint>
<constraint>
MUST NOT: Rewrite entire paragraphs without being asked.
</constraint>
<audience>
Target audience: intermediate general users

You can use technical terms but provide brief explanations when needed.
</audience>
<tone>
Tone: friendly
Be warm, approachable, and supportive.
Voice characteristics: formality: semi-formal
</tone>
<format>
Output format: markdown
</format>

Notice there's no extra constraints section at the end -- because you provided your own constraints, pupt skips the defaults.

What to Learn Next

You now know all the building blocks for structuring prompts. Next, make your prompts interactive:

• Variables & Inputs -- Collect user input and use it throughout your prompt
• Conditional Logic -- Include or exclude sections based on conditions
• Tags Overview -- Browse every tag available in pupt