Back to directory

LangGPT

A structured, reusable prompt-design framework, described as a programming language for prompts, that lets anyone build high-quality large-language-model prompts from hierarchical, template-based components.

LangGPT

LangGPT is a structured, reusable prompt-design framework that treats writing a prompt more like writing code than trial and error: prompts are organized hierarchically into templates that can be reused and adapted, with variables, commands and conditional logic available where needed, aimed at making prompt creation systematic rather than a collection of scattered tips.

Proposed in 2023, it became widely used within the Chinese AI community, to the point that many mainstream large language models, including ChatGPT, Claude, DeepSeek, Gemini and Qwen, respond to its keywords directly: simply asking a model to answer 'the LangGPT way' is often enough to get a structured, LangGPT-style prompt back, without writing out a template by hand. Dedicated generator tools, including a GPT-4-based generator and a version built for the Kimi model, can also produce a LangGPT-style prompt automatically from a plain description of what is needed.

The framework and its underlying paradigm are documented in an academic paper.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.