prompt-scribe 0.1.0

A powerful, template-based prompt composer for crafting and managing complex instructions for LLMs.

Prompt Scribe A powerful, template-based prompt composer for crafting and managing complex instructions for LLMs.

---

The Problem

Managing large, multi-part prompts for Large Language Models can be messy. You often combine a persona, rules, context from different files, and a specific query. Doing this manually by copy-pasting is tedious and error-prone.

The Solution

Prompt Scribe automates this process. It uses a simple YAML configuration and the powerful Jinja2 templating engine to compose your final prompts from various source files, ready to be used in any LLM chat or API.

Key Features

• Template-Based: Use Jinja2 templates for maximum flexibility.
• Modular: Split your prompts into reusable parts (personas, includes, rules).
• Safe Initialization: Creates a dedicated .prompt_scribe/ directory to avoid cluttering your project root.
• Watch Mode: Automatically recompose prompts when source files change.
• Beautiful CLI: A clean, helpful command-line interface.

Installation

pip install prompt-scribe

Quick Start

1. Initialize a new project: Navigate to your project directory and run:

   prompt-scribe init
   

   This will create a self-contained .prompt_scribe/ directory with a default structure:

   .
   └── .prompt_scribe/
       ├── composed_prompts/
       ├── includes/
       │   └── development-rules.md
       ├── personas/
       │   └── code-reviewer.md
       ├── templates/
       │   └── master.jinja2
       └── prompts.yml
   

2. Compose your prompts: Run the compose command from your project root:

   prompt-scribe compose
   

   This will read .prompt_scribe/prompts.yml, process the example-code-reviewer agent, and generate the final prompt in .prompt_scribe/composed_prompts/.

3. Use the output: Open the generated file, copy the content, and paste it into your LLM chat interface.

Usage

Composing Prompts

• Compose all agents:

  prompt-scribe compose
  

• Compose specific agents:

  prompt-scribe compose agent-one agent-two
  

• Watch for changes and automatically recompose:

  prompt-scribe compose --watch
  

Configuration (.prompt_scribe/prompts.yml)

The prompts.yml file is the heart of your project. It resolves all paths relative to its own location.

# Global settings for all agents
settings:
  personas_dir: "personas"
  includes_dir: "includes"
  templates_dir: "templates"
  output_dir: "composed_prompts"

# Map of agents to be composed
agents:
  example-code-reviewer:
    # Jinja2 template to use for rendering
    template: master.jinja2
    # The final output file name
    output_file: example_code_reviewer.md
    # The persona section, available as `persona` in the template
    persona:
      file: personas/code-reviewer.md
    # A list of content sections, available as `sections` in the template
    sections:
      - title: "📜 Key Development Rules"
        prologue: "These are the mandatory rules and principles for this project."
        file: includes/development-rules.md
      - title: "📄 Code Snippet for Review"
        # You can also include content directly
        content: |
          ```python
          def hello_world():
              print("Hello, Scribe!")
          ```

License

This project is licensed under the MIT License. See the LICENSE file for details.