wonderful-prompts: A Curated Chinese ChatGPT Prompt Library

What wonderful-prompts Is and Who It Serves

wonderful-prompts is a repository of curated Chinese-language prompts for conversational AI models, collected and refined by the author of the awesome-chatgpt-zh resource. Its goal is to help Chinese-speaking users get more predictable and useful outputs from ChatGPT and compatible models by providing tested, ready-to-use prompt templates rather than requiring each user to write prompts from scratch. The primary audience is Chinese-speaking engineers, content creators, students, and professionals who use ChatGPT or similar AI assistants regularly. The README notes that as of 2026, the prompts are applicable to current-generation models including GPT-5, Claude, DeepSeek-V4, and Gemini, not just the original ChatGPT. The repository is MIT licensed and links to a companion Feishu knowledge base with additional prompt material.

Repository Organization: Categories and What Each Covers

The repository organizes prompts into a table of contents with roughly twenty categories. The major divisions include: prompt generation and optimization (prompts that help users write better prompts), a general-purpose super-prompt, coding assistance (simulating a Linux terminal, writing functions, generating regular expressions, translating obfuscated code), writing (books, novels, xiaohongshu-style titles, grammar correction, style imitation), image and art generation (ASCII art, Midjourney and Stable Diffusion prompt generation), role-play, productivity, marketing strategy, personal development, and career planning. The coding section includes a resource called CAN (Code Anything Now), which is documented in examples/gpt4_CAN_coder.md. The examples/ directory contains fully worked-out prompt templates as Markdown files, covering topics like book generation, tutoring, and super-expert persona construction.

Using the Collection: Clone, Browse, and Copy

The repository requires no build step or runtime. Users clone or download it and read the README directly. Each prompt category links to its section in the document, and the prompts themselves are shown as code blocks that can be copied into a chat interface. The files/ directory holds additional resource files, and the imgs/ directory contains screenshot examples of the prompts in use. The examples/ directory holds complete prompt scripts in Markdown, including examples/Mr.-Ranedeer-AI-Tutor-zh, a detailed AI tutor persona prompt.

A typical entry is the "Fully unleash the power of o1" prompt, which instructs the model to use its full compute and token budget for a difficult question:

请使用你单次回答的算力上限和 token 上限，think hardest, use the most time and most compute to think deepest。
这是最深刻最复杂的问题，请给出你最高质量的回答。所以，你需要深度思考、独立思考、批判性思考、创造性思考。

For Stable Diffusion and Midjourney prompts, the README provides examples with specific style tags for 3D character modeling, environmental design, sculpture, VR and AR experiences, and children's book illustration. These prompts are copied into the respective image generation tools rather than used with ChatGPT.

The README also notes a technique for recovering from truncated AI responses: a short prompt that instructs the model to continue from exactly where it left off while maintaining the original format. This is documented explicitly as a section in the table of contents.

Guidance for Thinking-Mode Models

A 2026 note at the top of the README addresses how to write prompts for reasoning-capable models such as DeepSeek-V4 in thinking mode, Claude with extended thinking, and GPT-5 in reasoning mode. The guidance differs from standard prompt engineering advice. For these models, the README recommends focusing on stating the goal and acceptance criteria clearly, reducing or eliminating few-shot examples, and avoiding explicit instructions like "think step by step" that were useful for earlier models but are unnecessary with current reasoning models. The README links to two companion repositories, awesome-deepseek-prompts and awesome-claude-prompts, for model-specific prompt collections.

This makes the repository more useful in 2026 than a static historical collection would be, since the author actively updates the guidance as the AI model landscape changes. The practical implication is that prompts originally written for GPT-3.5 may need adjustment when used with a thinking-mode model, and the README acknowledges this directly.

Limitations: Chinese-First Presentation and No Structured API

The primary limitation of wonderful-prompts is that all prompts, section headings, and explanatory text are in Chinese. For engineers who do not read Chinese, the prompts require translation before use, which removes much of the convenience the repository is designed to provide. The related awesome-chatgpt-prompts repository by the user f/awesome-chatgpt-prompts is a similar collection focused on English-language prompts.

The repository has no programmatic interface or searchable prompt database. Browsing is done by reading the README in a web browser or text editor, navigating by the table of contents. There is no tagging system, similarity search, or metadata format that would allow automated retrieval of prompts by category or use case. For a team that wants to integrate prompts into a pipeline, this collection requires manual curation. The repository also has no mechanism for testing whether a given prompt still produces the intended output after a model version change.

wonderful-prompts versus awesome-chatgpt-prompts

The f/awesome-chatgpt-prompts repository on GitHub is the most widely referenced comparable collection for English-speaking users. It provides prompts organized by persona ("Act as..." style) and was one of the first large-scale prompt collections published after ChatGPT's release. Like wonderful-prompts, it is a static Markdown document with no API, no build step, and an MIT license.

The practical difference is language and editorial focus. awesome-chatgpt-prompts is organized around persona prompts that instruct the model to act as a specific type of expert, while wonderful-prompts covers a wider range of task types, including image generation, coding tools, and platform-specific writing styles like Xiaohongshu. wonderful-prompts is actively updated and includes guidance for 2026 reasoning models, whereas awesome-chatgpt-prompts has a larger contribution history but focuses on the original ChatGPT prompt pattern. For Chinese-speaking users, wonderful-prompts is a more directly applicable resource.

Editorial conclusion

wonderful-prompts suits Chinese-speaking engineers and content creators who want a ready-made library of ChatGPT prompts they can copy, adapt, and extend without starting from scratch. For teams or individuals who primarily work in English, the Chinese-first presentation is a friction point, though the prompts themselves are intelligible with translation. The MIT license permits redistribution and modification without restriction. The companion LangGPT framework at github.com/yzfly/LangGPT is the recommended tool for writing new structured prompts in the same style.

Frequently asked questions

Official sources

1. Issues
2. langgptai/wonderful-prompts on GitHub
3. License: MIT
4. Project website
5. README