> ## Documentation Index
> Fetch the complete documentation index at: https://prompt-lockbox.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# AI-Based Prompt Iteration

> Transform rough prompt ideas into polished, versioned assets using AI-powered iteration.

<Tip>
  **Got any cool idea?**

  Share your idea via [Suggest Feature](/sf) or contribute directly—learn how at [Contribution Guide](/contri).
</Tip>

Great prompts are rarely written perfectly on the first try. They evolve through a cycle of **drafting**, **testing**, and **refining**. Prompt Lockbox's AI features are designed to supercharge this cycle, turning a slow, manual process into a rapid, intelligent workflow.

This guide will walk you through a complete iteration loop. We'll start with a rough, undocumented prompt idea and, in just a few commands, **use AI to transform it into a documented, improved, and versioned asset ready for use.**

<Info>
  **The Goal**: To demonstrate a rapid development workflow:

  Start with a basic prompt template.

  * Use AI Documentation to create a baseline understanding.
  * Use AI Improvement to refine and strengthen the prompt.
  * Version the final, polished result.
</Info>

Here are the steps:

<Steps>
  <Step title="Create the Initial Draft">
    First, **let's create a new prompt**. Imagine we need a prompt to summarize meeting transcripts. We'll start by creating a file with a name and a simple template, **but leave the description and tags blank**.

    In your teminal, run the command:

    ```bash theme={null}
    plb create
    # Name: meeting-summarizer
    # (Leave description and tags blank)
    ```

    Created prompt file:

    ```yml prompts/meeting-summarizer.v1.0.0.yml theme={null}
    name: meeting-summarizer
    version: "1.0.0"
    description: ""
    tags: []
    ... 
    template: |
      Here is a meeting transcript. Please summarize it.

      TRANSCRIPT:
      ${transcript_text}
    ```

    This prompt is functional, but it's vague and undocumented.
  </Step>

  <Step title="Generate Documentation with AI">
    Instead of manually writing a description and thinking of tags, let's use the `document` command.

    <CodeGroup>
      ```bash CLI theme={null}
      plb prompt document meeting-summarizer
      ```

      ```python Python theme={null}
      from prompt_lockbox import Project

      project = Project()
      prompt = project.get_prompt("meeting-summarizer")
      if prompt:
          prompt.document()
      ```
    </CodeGroup>

    **Result**: The AI analyzes the template and automatically updates the `.yml` file:

    ```yml prompts/meeting-summarizer.v1.0.0.yml theme={null}
    ... 
    description: "Summarizes a provided meeting transcript text into a concise overview."
    tags: [summarization, meetings, transcription, nlp]
    template: |
      Here is a meeting transcript. Please summarize it.
      
      TRANSCRIPT:
      ${transcript_text}
    ```

    Now our prompt is documented and discoverable, but the template itself can still be improved.
  </Step>

  <Step title="Improve the Prompt with AI">
    Our current template is too generic. Let's ask the AI to make it more specific and structured. Optionally, **you can provide notes to LLM using `note`.**

    <Note>
      You can use the CLI or Python, but the **CLI is recommended** for clearer visibility into the changes.
    </Note>

    <CodeGroup>
      ```bash CLI theme={null}
        plb prompt improve meeting-summarizer --note "Make it output action items and decisions separately."
      ```

      ```python Python theme={null}
      # ... (get the prompt object as before) ...
      if prompt:
          critique = prompt.get_critique(
              note="Make it output action items and decisions separately."
          )
          # Apply the improvement
          prompt.improve(critique["improved_template"])
      ```
    </CodeGroup>

    **Result**: The `improve` command will suggest a much more robust template. After you confirm the changes, your file will be updated. Here's the updated file:

    <CodeGroup>
      ```yml prompts/meeting-summarizer.v1.0.0.yml theme={null}
        template: |
          You are an expert meeting assistant. Analyze the following meeting transcript and provide a structured summary.
        
          Your summary must contain two distinct sections:
          1.  **Key Decisions:** A bulleted list of all decisions made during the meeting.
          2.  **Action Items:** A bulleted list of all tasks assigned, including who is responsible if mentioned.
          
          If no decisions or action items are found, state "None".
        
          TRANSCRIPT:
          ${transcript_text}
      ```
    </CodeGroup>
  </Step>

  <Step title="Version the Polished Prompt">
    We've transformed our rough draft into a documented and structured prompt. The final step is to lock this polished version or create a new one to signify that it's a major improvement.

    ```bash theme={null}
    # We can lock v1.0.0 now that it's complete
    plb lock meeting-summarizer

    # Or, if we feel the change was significant, create v1.1.0
    plb version meeting-summarizer --minor
    ```
  </Step>
</Steps>

This rapid iteration loop — **Create -> Document -> Improve -> Version** —is at the heart of what makes the toolkit useful.