repository·main·Indexed 11 days ago
https://github.com/langgptai/langgptA systematic, template-based methodology for creating high-quality LLM prompts, treating prompt engineering like a programming language. It utilizes a role-based framework with modular components such as Role, Profile, Goal, Skills, Rules, and Workflow. The ecosystem includes tools like PromptShow for creating prompt images, PromptVer for semantic versioning, and specialized meta-prompts for automated prompt generation.
The LangGPT ecosystem consists of several specialized tools and collections:
Curated prompts optimized for specific models:
wonderful-promptsawesome-claude-promptsawesome-deepseek-promptsawesome-gemini-promptsawesome-grok-promptsqwen-promptsawesome-llama-promptsawesome-doubao-promptsFor inquiries or to join the community, use the following resources:
A Structured Prompt is a method of writing prompts using organizational patterns similar to writing an article or using templates (like a CV or a research paper). Instead of providing a single block of unstructured text, you use a combination of Identifiers and Attribute Words to define the semantic structure of the prompt.
#, ##, <>, [], or - used to denote hierarchy, headings, or variables. LangGPT primarily uses Markdown syntax (# for H1, ## for H2, etc.) to control content levels.Role, Profile, Rules, Workflow, and Initialization. These act like section headings in an academic paper (e.g., Abstract, Methods, Conclusion) to tell the model what the following content describes.While frameworks like CRISPE (Capacity, Insight, Statement, Personality, Experiment) provide a mental model for what to include, Structured Prompting provides a formal template for how to present that information to the LLM.
Relationship Expression is a prompting strategy designed to overcome the limitations of using technical terminology (Entity Expression). Instead of trying to 'compress' complex feelings into precise terms like 'Minimalism' or 'Material Design'—which can lead to ambiguity or information loss—you 'expand' your requirements by describing the relationships between known and unknown things.
This method is highly effective for AI because LLMs operate on vector mathematics, understanding concepts based on their position in a 'relationship space' relative to other concepts.
LangGPT (Language for GPT-like LLMs) is a natural language programming framework designed to improve the quality, reusability, and structure of prompts for Large Language Models (LLMs).
Inspired by the structure and reusability of programming languages, LangGPT moves away from unstructured natural language instructions toward a structured, two-layer design consisting of modules and internal elements. This approach aims to reduce the learning curve for non-AI experts and allow LLMs to automatically generate high-quality prompts in a manner similar to code generation.
A Prompt is any content used by a Large Language Model (LLM) to predict an output. This includes:
Core Concept: "Everything that is used by the model to predict an output result is a Prompt."
Prompt Engineering is the practice of designing and adjusting prompts to obtain ideal model results. It is the interface used to call and optimize the capabilities of an LLM. It is a foundational step in LLM application development, serving as the basis for performance optimization, RAG (Retrieval-Augmented Generation), and fine-tuning data construction.
To optimize prompts, LangGPT follows four core design principles derived from the comparison between natural language and programming languages:
Structured Prompts offer several key advantages over unstructured text:
# Role, ## Profile, ### Skill) to unify form and content, improving readability for humans and aligning with the hierarchical training data of LLMs like ChatGPT.Role, Expert, or Master directly triggers the model's role-playing capabilities. Using Rules or Constraints helps mitigate hallucinations and ensure safety.Input and Output modules for multi-agent chains).The essence of prompting is a cognitive practice rather than a mere technical skill. It is designed to teach four core abilities:
Core Principle: AI can generate content, but it cannot confer meaning. Meaning remains the exclusive domain of human intention, choice, and interpretation.
Enhance prompt intelligence using these advanced syntax patterns:
1. Variables and References
Use <Variable> syntax to create self-referential prompts that maintain consistency.
Example: As a <Role>, you must follow <Rules>.
2. Commands
Define reusable actions for better UX, typically using a / prefix.
Example:
## Commands
- Prefix: "/"
- Commands:
- help: Display all available commands
- continue: Resume interrupted output3. Conditional Logic
Use If/Else structures to handle different user inputs.
Example:
If user provides [code], then analyze and suggest improvements
Else if user asks [question], then provide detailed explanation
Else, prompt for clarification4. Reminders
Use a ## Reminder section to combat context loss in long conversations by explicitly restating active rules or language settings.