langgptai/wonderful-prompts: A Chinese Prompt Collection for ChatGPT and Reasoning Models

What langgptai/wonderful-prompts is for

The repository is a curated list of Chinese-language prompts for ChatGPT and comparable assistants. The README describes it as "中文 prompts 精选" and says it contains hundreds of high-quality prompts with illustrated usage examples, so a reader can learn how to use ChatGPT better. The audience is therefore not a platform team building an LLM pipeline. It is an individual user, a writer, a developer, or a marketer who wants a prompt they can paste into a chat window today.

The stated lineage matters. The README says the project is a series of Chinese ChatGPT prompts selected and refined by the author of the ChatGPT Chinese guide, and it points to two sibling projects: awesome-chatgpt-zh for tutorials and curated open source projects, and LangGPT for writing high-quality ChatGPT prompts. That positions wonderful-prompts as the copy-paste layer, with LangGPT as the authoring method and awesome-chatgpt-zh as the broader reading list. If you want a methodology, this is not it. If you want a prompt for writing a weekly report, naming a product, or generating Midjourney spells, it is.

How the repository is organised and how prompts flow into a chat

The top-level layout is flat: LICENSE, README.md, examples/, files/, and imgs/. The README itself carries the table of contents and the prompt text, grouped by task. The categories visible in that table of contents include prompt generation and optimisation (Prompt 工程师, Prompt 工程专家, Meta Prompt, Prompt 评分专家), a general super prompt, coding (资深编程专家 CAN, 编写正则表达式, 模拟 Linux 终端), writing (撰写一本书籍, 小红书爆款标题生成器, 将文本转换为表格), drawing and images (绘制 ASCII 字符画, ChatGPT 生成 Midjourney 咒语), and a long tail of role-play and productivity prompts such as 周报生成器, 简历生成器, and 私人订制健身计划.

The data flow is manual and that is the whole mechanism. You open the README, find the section, copy the fenced prompt block, and paste it into ChatGPT or another assistant. There is no server, no prompt registry, no templating layer, and no API call in the repository. The examples/ directory holds longer worked examples, with files such as examples/gpt4_CAN_coder.md, examples/gpt_math_solver.md, examples/super_experts_gpt.md, examples/GPT_Generate_A_book.md, and examples/base.md. Those are Markdown documents, so the same copy-and-paste applies.

One structural detail is worth noting for anyone who plans to reuse the text programmatically: the prompts live inside fenced code blocks in the README, sometimes with a leading quote crediting an external source, as with the o1 prompt that the README attributes to a post on X. There is no machine-readable index, so extracting prompts at scale means parsing Markdown yourself.

Installing nothing: first use of a wonderful-prompts prompt

There is no installation step, because the project ships no executable code. To get the files locally, clone the repository:

git clone https://github.com/langgptai/wonderful-prompts.git
cd wonderful-prompts

After the clone you get the README, the examples directory, the files directory, and the images directory. The README is the entry point; open it and use the table of contents to jump to a category.

A first real use is the weekly report prompt, listed in the README as 周报生成器. Copy the prompt block from the README into the chat input, then append your own raw notes for the week. The README's own framing is that the prompt turns scattered work notes into a structured report; the model returns the formatted result in the conversation, and you edit from there.

If you want to work from a longer example instead, read one of the Markdown files in examples/ before pasting. For instance:

ls examples/

The listing shows the example set, including gpt4_CAN_coder.md and gpt_math_solver.md. Those files show the prompt together with how it is meant to be used, which is more useful than the bare prompt when the prompt expects a specific input format.

The reasoning-model caveat in the README

The README carries a 2026 note stating that the repository's prompts are general across contemporary models including GPT-5, Claude, DeepSeek-V4 and Gemini. It then gives a concrete piece of advice for thinking models: when writing prompts for DeepSeek-V4, Claude extended thinking, or GPT-5 reasoning mode, state the goal and the acceptance criteria clearly, use fewer few-shot examples, and drop the instruction to think step by step. It links to a DeepSeek V4 thinking-mode section in a sibling guide and to two further repositories, awesome-deepseek-prompts and awesome-claude-prompts.

This is the most useful editorial content in the repository, and it is also a warning about the collection itself. Many of the prompts were written for older chat models, where chain-of-thought phrasing and heavy few-shot scaffolding helped. On a reasoning model those same instructions can be redundant or actively unhelpful. The README acknowledges the shift but does not annotate which individual prompts predate it. A reader therefore has to judge per prompt whether the phrasing still fits the model in front of them. That is a real cost of a curated list that grows by accumulation.

Where the collection falls short

The main limitation is that a prompt list is not a maintained interface. Nothing in the repository tells you which prompts were validated against which model version, and recent releases were not retrieved, so there is no changelog to consult for that either. The README does say the project keeps updating and invites issues with interesting prompts, and the last push was on 2026-08-25, so the repository is not dormant. Activity, however, is not the same as per-prompt validation.

A second limitation is language. The prompts and the surrounding documentation are in Chinese. If your team works in English, the prompts are still readable but you will be translating and re-testing them, at which point you have lost most of the convenience the collection offers.

A third is scope. The README lists prompts for entertainment and role-play alongside professional ones, including 红颜知己, 可爱女友, 夸夸怪, 吵架小能手, and 抬杠高手. That is part of the project's stated goal of making ChatGPT more playable, but it means the repository is not a vetted library of production prompts. If you need prompts with owners, review, and regression tests, this is the wrong tool, and the repository does not claim otherwise.

How it differs from LangGPT and awesome-chatgpt-zh

The README names two related projects, and the difference in approach is explicit rather than implied. LangGPT is presented as the way to write high-quality ChatGPT prompts, meaning it is a structured method for authoring prompts, with its own conventions for defining roles, goals, and constraints. wonderful-prompts is the output side: a set of finished Chinese prompts you consume directly. If your problem is "I do not know how to phrase this instruction," LangGPT is the relevant project. If your problem is "I need a prompt for this task right now," wonderful-prompts is.

awesome-chatgpt-zh sits one level further out, as a general Chinese guide to ChatGPT with tutorials and curated projects. It is reading material rather than a prompt shelf. The practical consequence is that wonderful-prompts is the shallowest of the three by design, and that is fine as long as you treat it as a starting point. The repository's own structure supports that reading: prompts in the README, worked examples in examples/, and links out to deeper guides when you want to understand why a prompt is shaped the way it is.

Licence and the cost of keeping up

The repository is licensed MIT, and the LICENSE file sits at the top level. MIT permits reuse, modification, and redistribution with the licence and copyright notice retained, which makes it practical to copy individual prompts into internal documentation or a product. This is a description of the licence text, not legal advice; if you plan to redistribute the collection or embed prompts in a commercial product, check the notice requirements with your own counsel.

Upgrade cost is low in the mechanical sense, because there is nothing to upgrade: pull the repository again and diff the README. The real cost is revalidation. Each time you move to a new model generation, the prompts that used chain-of-thought phrasing or heavy examples may need rewriting, and the README's own 2026 note is evidence that this already happened once. Budget for reading the diff and re-testing the handful of prompts you actually depend on, rather than assuming a fresh clone keeps your results stable.

Editorial conclusion

Adopt langgptai/wonderful-prompts if you write Chinese prompts for general assistants and want a curated starting point you can copy into a chat window; the repository is a text collection, so there is nothing to install and nothing to break. Do not adopt it if you need versioned prompt artifacts, automated evaluation, or English-language prompts, since the README does not describe a CLI, an SDK, or a test harness. Before relying on it, check whether the specific prompt you want still matches the model you use, because the README notes that its prompts target GPT-5, Claude, DeepSeek-V4 and Gemini, and the reasoning-model guidance differs from the older few-shot style.

Frequently asked questions

Official sources

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