Patterns and Prompting

Fabric's core innovation lies in its concept of Patterns, which are fundamentally well-structured AI prompts designed to augment human capabilities by solving specific real-world tasks. Instead of focusing on AI as a standalone entity, Fabric prioritizes the "human problem" it aims to solve, breaking down complex challenges into manageable components that AI can assist with. This approach addresses what Fabric identifies as AI's primary challenge: not a capabilities problem, but an integration problem. Patterns serve as organized, reusable units of AI intelligence, allowing users to collect, create, and deploy their most effective AI solutions across various tools and interfaces.

Fabric's Approach to Prompting

Fabric's prompting methodology is distinct and emphasizes clarity, readability, and modularity. Unlike many traditional prompts, Fabric Patterns are designed with several key principles:

1. Markdown for Readability and Editability: All Fabric Patterns are written in Markdown. This choice is deliberate, ensuring maximum readability and editability for both human users and the AI models themselves. The structured nature of Markdown helps to clearly delineate sections and instructions, making it easier for creators to craft effective prompts and for users to understand and modify them.
2. Clear and Explicit Instructions: Patterns are crafted with exceptionally clear and explicit instructions. Markdown structures like headings, bullet points, and emphasis are utilized to highlight what the AI needs to do and in what specific order. This precision reduces ambiguity and improves the consistency and quality of AI responses.
3. Predominant Use of the System Section: Fabric primarily leverages the "System" section of a prompt. Extensive testing has shown that placing the core instructions and constraints within the AI's system message often leads to more effective and reliable behavior from the language model. This ensures the AI understands its role and the task before processing user input.

Structure of a Fabric Pattern

A typical Fabric Pattern follows a consistent Markdown-based structure, designed for clarity and comprehensive instruction. While the exact content varies by task, the foundational sections are usually:

Let's break down these sections:

• # IDENTITY and PURPOSE: This section defines the AI's persona and its overarching goal. It establishes the context and role the AI should adopt (e.g., "You are an expert summarizer," or "You are a creative writer assisting with story ideas.").
• # STEPS: This section outlines the sequence of actions or thought processes the AI should follow to achieve the stated purpose. These are usually presented as numbered or bulleted lists, guiding the AI through the task step-by-step.
• # OUTPUT INSTRUCTIONS: This section specifies the desired format, length, tone, and any other constraints for the AI's final output. It ensures the response is delivered in a usable and consistent manner, often including examples or specific formatting requirements.

You can explore a wide variety of pre-built patterns in the data/patterns directory of the Fabric repository, or create your own as detailed in the Custom Patterns page.

Prompt Strategies

Beyond individual Patterns, Fabric introduces Prompt Strategies to further refine and enhance AI interactions. These strategies are meta-prompts that modify the behavior of a basic pattern, enabling more advanced reasoning techniques. Examples include "Chain of Thought" or "Chain of Draft," which guide the AI through a series of internal steps before producing a final output.

Each strategy is defined in a small JSON file. These files are located in the data/strategies directory within the Fabric repository, and once installed, they reside in your ~/.config/fabric/strategies directory.

To install these strategies, you can use the Fabric setup command:

During the setup process, you'll be prompted to select the "Strategies" option from the "Tools" menu, which will download and install the available strategies into your configuration directory. Once installed, you can leverage these strategies to modify the behavior of your chosen patterns, unlocking more sophisticated AI capabilities. For more information on how to utilize these strategies in your commands, refer to the CLI Usage Guide and Advanced Features pages.