MindStudio Generate Text Block Prompting Skill#

A skill for writing production-ready prompts for MindStudio's Generate Text block, covering text output, JSON output, variable injection, conditional logic, and dot-notation data access.

---

Step 1: Interview the User First#

Always gather context before writing a single line of prompt. The quality of a MindStudio prompt depends entirely on knowing what variables exist, what the block is supposed to do, and what comes next in the workflow. Never guess at variable names — use the exact names the user has defined.

Ask the user these questions before starting. You don't need all answers for every case, but get as many as apply:

About the block's purpose:

• What should this Generate Text block do? (What's the job?)
• Where does it sit in the workflow — early, middle, or final step?

About the inputs (variables coming IN):

• What variables are available at this point in the workflow?
• What are their exact names (e.g., topic, userMessage, scrapedContent)?
• Are any of them JSON objects or arrays? If so, what does the structure look like?

About the output:

• Should the output be shown to the user, or saved to a variable for later use?
• If saved to a variable, what will you name it?
• Should the output be plain text, or structured JSON?
• If JSON, what fields does the next block need to access?

About what comes after:

• Does a Run Workflow block or another Generate Text block consume this output?
• If iterating, what does each item in the array represent?

Once you have these answers, use the exact variable names the user provided throughout the prompt — never invent placeholder names like {{yourVariable}} or {{inputData}}.

---

How the Generate Text Block Works#

The Generate Text block sends a prompt to an AI model and returns a response. There are two core decisions to make before writing any prompt:

1. Output Behavior

• Display to User — the response is shown directly to the end user.
• Assign to Variable — the response is saved to a named variable (e.g., reportJSON) for use in later blocks.

2. Output Schema

• Text (Default) — plain text or markdown. Good for display, emails, chat responses.
• JSON — structured data. Required when downstream blocks need to access specific fields, iterate over arrays, or pass data to sub-workflows.
• CSV — tabular data for spreadsheets.

Always confirm both before writing the prompt. The output schema shapes every other decision.

---

Prompt Structure Best Practices#

MindStudio prompts are plain text written directly into the block's Prompt field. Use Markdown headings and sections to organize longer prompts — the AI reads and follows structure well.

The Basic Template#

text

<context_variable>
  {{variableName}}
</context_variable>

## Task
Describe exactly what the AI should do.

## Style
Describe tone, length, and format expectations.

## Output
Describe what the final response should look like.

Use XML-style tags (e.g., <topic>, <research>, <userInput>) to wrap injected variables. This clearly separates data from instructions and reduces hallucination.

Injecting Variables#

Use double curly braces to inject any workflow variable into the prompt:

text

The user's topic is: {{topic}}
Today's date is: {{currentDate}}

Variables are resolved at runtime — whatever value the variable holds at that moment gets injected as plain text.

---

Text Output Prompts#

For Text (Default) output, write the prompt as a clear set of instructions. Be specific about format and length.

Example: Simple Text Output#

text

<userInput>
  {{userMessage}}
</userInput>

Respond to the user's message above in a friendly, helpful tone.
Keep your response to 2-3 paragraphs.
Do not include headers or bullet points.

Example: Markdown-Formatted Report#

text

<topic>
  {{topic}}
</topic>
<researchNotes>
  {{allResearchMaterials}}
</researchNotes>

## Task
Write a comprehensive report based on the research notes above.

## Formatting Rules
- Use ## for major section headings
- Use ### for subsection headings
- Write in prose paragraphs, not bullet points
- Minimum 800 words

## Style
Write like a Bloomberg or NYT analyst: specific, factual, engaging.

Markdown Reference for Prompts#

Use these inside prompts to control AI output formatting:

ElementSyntaxUse it forHeading 1# TitleTop-level report titleHeading 2## SectionMajor sectionsHeading 3### SubsectionSubsectionsBold**text**Emphasis on key instructionsItalic*text*Secondary emphasisBullet list- itemUnordered listsNumbered list1. itemOrdered stepsBlockquote> textCallouts or notesCode`code`Variable names or exact strings

---

JSON Output Prompts#

When Output Schema is set to JSON, the prompt must instruct the model to return only valid JSON and nothing else — no preamble, no markdown fences, no explanation.

The Golden Rule for JSON Prompts#

Always end with an explicit output instruction that shows the exact schema:

text

Respond only with valid JSON. Do not include any explanation, preamble, or markdown code fences.

JSON Output:
{
  "key": "value"
}

The Sample Output Field (MindStudio Block Setting)#

When Output Schema is set to JSON in MindStudio, a Sample Output field appears below it. This is separate from the prompt — it tells MindStudio what the JSON structure looks like so it can parse and route the data correctly.

The Sample Output must use realistic example data, not placeholder types like "string" or "number. MindStudio reads this field to understand the shape of the output, so it needs to look like a real response the AI would actually return.

For every JSON prompt you write, always deliver two things:

1. The prompt itself (with the schema at the bottom using type hints like "string")
2. A separate Sample Output block with realistic fake data to paste into the MindStudio block setting

Example — People Extraction:

Prompt schema (inside the prompt):

json

[
  {
    "firstName": "string",
    "lastName": "string",
    "jobTitle": "string",
    "company": "string",
    "linkedin": "string"
  }
]

Sample Output (paste into the MindStudio block setting):

json

[
  {
    "firstName": "Sarah",
    "lastName": "Chen",
    "jobTitle": "VP of Marketing",
    "company": "Acme Corp",
    "linkedin": "https://www.linkedin.com/in/sarahchen"
  }
]

Example — Research Report:

Prompt schema (inside the prompt):

json

{
  "title": "string",
  "subtitle": "string",
  "sections": [
    {
      "header": "string",
      "subsections": [
        {
          "h3": "string",
          "paragraphs": ["string"]
        }
      ]
    }
  ]
}

Sample Output (paste into the MindStudio block setting):

json

{
  "title": "The Rise of Autonomous AI Agents",
  "subtitle": "How software that acts on its own is reshaping the enterprise",
  "sections": [
    {
      "header": "Market Overview",
      "subsections": [
        {
          "h3": "Current Adoption",
          "paragraphs": ["Enterprise adoption of AI agents has accelerated rapidly since 2023, with over 40% of Fortune 500 companies piloting autonomous workflow tools."]
        }
      ]
    }
  ]
}

Rules for writing Sample Output:

• Use realistic names, titles, dates, and sentences — not "string", "value", or "example"
• Match the exact same keys and nesting structure as the prompt schema
• For arrays, include just one item — MindStudio only needs to see the shape
• For strings, write a realistic short value (a real name, a real sentence, a real URL)

Example: Simple JSON Output#

text

<userTopic>
  {{topic}}
</userTopic>

Generate 3 diverse Google search queries that would help research the topic above.

Respond only with a JSON array of strings. No explanation, no markdown.

JSON Output:
["query1", "query2", "query3"]

Example: Complex Nested JSON Output#

text

<articleMaterials>
  {{allResearchMaterials}}
</articleMaterials>

Write a full research report based on the materials above.

## Output Rules
- Title: concise, descriptive
- Subtitle: one compelling hook sentence
- At least 3 sections, each with 2+ subsections
- Each subsection must have at least 2 long paragraphs
- Sources: include URL and title for each source used

Respond only with valid JSON matching this exact schema. No markdown, no preamble.

JSON Output:
{
  "title": "string",
  "subtitle": "string",
  "sections": [
    {
      "header": "string",
      "subtitle": "string",
      "subsections": [
        {
          "h3": "string",
          "paragraphs": ["paragraph 1", "paragraph 2"]
        }
      ]
    }
  ],
  "sources": [
    {
      "url": "string",
      "title": "string"
    }
  ]
}

Common JSON Output Mistakes to Avoid#

MistakeFixAI wraps output in ```json fencesAdd "Do not use markdown code fences" to promptAI adds explanation before the JSONAdd "Respond ONLY with JSON and nothing else"AI uses single quotesSpecify "Use double quotes for all keys and values"AI adds trailing commasAdd "Ensure valid JSON with no trailing commas"Output schema is missingAlways include a sample JSON schema in the prompt

---

Accessing JSON Data with Dot Notation#

When a variable holds a JSON object or array, access specific fields using dot notation inside any subsequent prompt or block.

Accessing Object Properties#

If a variable user holds:

json

{
  "name": "Alice",
  "contact": {
    "email": "alice@example.com"
  }
}

Access it in a prompt like this:

text

Name: {{user.name}}
Email: {{user.contact.email}}

Accessing Array Items#

If a variable data holds:

json

{
  "fruits": ["apple", "banana", "cherry"]
}

Access by index (zero-based):

text

First fruit: {{data.fruits.[0]}}
Second fruit: {{data.fruits.[1]}}

Accessing Arrays of Objects#

If a variable team holds:

json

{
  "users": [
    { "id": 1, "name": "Alice" },
    { "id": 2, "name": "Bob" }
  ]
}

Access like this:

text

First user: {{team.users.[0].name}}
Second user's ID: {{team.users.[1].id}}

Real-World Pattern: Sub-workflow Receives a Person Object#

In a sub-workflow launched with a person launch variable:

json

{
  "firstName": "Mark",
  "lastName": "Benioff",
  "jobTitle": "CEO",
  "company": "Salesforce"
}

Use in the prompt:

text

Find the work email for {{person.firstName}} {{person.lastName}},
who is the {{person.jobTitle}} at {{person.company}}.

---

Conditional Logic in Prompts#

Use {{#if}} / {{else}} / {{/if}} to branch prompt behavior based on whether a variable exists and has a value.

Basic Conditional#

text

{{#if customerName}}
Write a personalized thank-you email to {{customerName}}.
{{else}}
Write a general thank-you email to all customers.
{{/if}}

Nested Conditionals#

text

{{#if userType}}
  {{#if isPremium}}
  Generate a premium onboarding message for {{userType}}.
  {{else}}
  Generate a standard onboarding message for {{userType}}.
  {{/if}}
{{else}}
Generate a generic onboarding message.
{{/if}}

When to Use Conditionals#

• When a variable may or may not exist depending on the workflow path
• When the prompt should behave differently for different user types
• When an optional field (like a name or preference) should personalize the output if present

---

Run Workflow + JSON Array Pattern#

When a Generate Text block outputs a JSON array and a downstream Run Workflow block iterates over it, each item becomes {{item}} in the sub-workflow.

Parent Workflow Prompt (generates the array)#

text

Generate an array of 3 Google search queries for the topic: {{topic}}

Respond only with a JSON array. No explanation.

JSON Output:
["query1", "query2", "query3"]

Configure the Run Workflow block:

• Iterator: JSON Array Input (Advanced)
• Input Array: {{queries}}
• Launch Variable: query : {{item}}

Sub-workflow Prompt (receives one item)#

text

Search query: {{query}}

Based on this query, write 3 key findings from research on the topic.

---

Quick Checklist Before Finalizing a Prompt#

For any prompt:

• Are all variables wrapped in {{double curly braces}}?
• Is injected data wrapped in descriptive XML tags (e.g., <userInput>, <research>)?
• Is the task instruction clear and specific?
• Is the expected output format stated explicitly?

For JSON output:

• Does the prompt say "Respond ONLY with valid JSON"?
• Does the prompt say "No markdown, no preamble, no code fences"?
• Is the full JSON schema shown as a sample at the end of the prompt?
• Is the Output Schema in the block settings set to JSON?
• Has a Sample Output been provided with realistic example data (not "string" placeholders) to paste into the MindStudio block setting?

For dot notation access:

• Is the variable name correct (matches the Output Variable name from the block that created it)?
• Are array indexes wrapped in square brackets: .[0], .[1]?
• Is each property access separated by a .?