A Chinese translation of the AI coding tool system prompts

A translation mirror, not a tool you can run

The first thing to understand is that this repository contains no software. It is a Chinese translation of an English project, system-prompts-and-models-of-ai-tools, and the stated purpose is to give Chinese developers and AI enthusiasts the system prompts and model design documents of popular AI programming tools, so they can understand how the assistants work and interact with them more effectively. The repository description adds that it covers Cursor, Devin and VSCode Agent among others, and that it is continuously updated with Chinese programming Rules and the latest AI programming prompts. The badges at the top show a version of 1.0.0, the MIT licence and an invitation to contribute. There is a DISCLAIMER.md in the tree and a statement that the project is for study and research only, with all the prompts and model documents there to help developers understand how the AI tools work internally. For a reader who works in Chinese, the appeal is straightforward. The same material exists in English upstream, but a translated copy of a technical instruction document is not merely a convenience, because the reasoning inside these prompts is dense, repetitive and written to be read by a model rather than by a person, and reading them in a second language costs a great deal more attention.

The directory listing is the actual content

The structure of the repository is one directory per tool, and the top level list is the quickest way to see the scope. The directories include Amp, Anthropic, Augment Code, Claude Code, Cluely, CodeBuddy Prompts, Comet Assistant, Cursor Prompts, Devin AI, Emergent, Google, Junie, Kiro, Leap.new, Lovable, Manus Agent Tools & Prompt, NotionAi, Open Source prompts, Orchids.app, Perplexity, Poke, Qoder, REA Creator, Replit, Same.dev, Trae, Traycer AI, VSCode Agent, Warp.dev, Windsurf, Xcode, Z.ai Code, dia, and v0 Prompts and Tools. What the README's own tree section shows for a smaller set of them describes the pattern: the Cursor directory holds Cursor editor prompts, Devin AI holds Devin system prompts, VSCode Agent holds related documentation, Replit holds prompts and tools, Windsurf holds related documents, Same.dev holds platform prompts, Lovable holds AI assistant documents, Manus Agent Tools & Prompt holds tools and prompts, Open Source prompts holds prompts for open source AI tools, REA Creator holds reverse engineering prompts, and v0 Prompts and Tools holds v0's prompts and tools. Two of those groupings are worth noting for what they reveal about the corpus. The Open Source prompts directory collects prompts for open source AI tools, and the REA Creator entry covers reverse engineering, which is a different category of instruction entirely from writing application code.

Four reasons to read them, according to the project

The README lists what these prompts and model design documents are for, and the list is a good description of the use cases. The first is understanding the internal working mechanism of AI programming tools, which is the primary motivation and the one the disclaimer reinforces. The second is improving how you interact with those tools, which is a practical use: if a prompt shows that the assistant is told to do something in a particular order, or to prefer a particular tool, you can work with that rather than against it. The third is providing a reference for building similar tools, which is how an engineer writing their own agent can see how a commercial one is structured without reverse engineering a product. The fourth is learning the best practices of AI agent design, which is the framing that makes this repository useful even when the specific tools change, because the patterns of how an agent is instructed outlive any one vendor's version. The planned future content follows the same logic, with Chinese programming Rules aimed at giving Chinese developers programming rules and guidelines that suit them better, more tool prompts as the upstream set changes, prompt optimisation advice for Chinese users, and case studies of the prompts working in real development.

The disclaimer and the licence situation

This is the part that deserves the most careful reading before anyone builds anything on top of the collection. The repository itself states that it is for study and research use only, and that the prompts and model documents are there to help developers understand how the AI tools work, with the detailed disclaimer in DISCLAIMER.md, which is present in the tree. The repository licence is MIT, and the README says so in Chinese and points at the LICENSE file. Here is the tension. The MIT licence covers the repository's own contribution, which is the translation work, but the material being translated is the system prompt of a commercial product, and those texts are not the translator's to license. The original English project exists because those prompts have been published and collected; the translation inherits their status rather than changing it. The repository is explicit that the original source project is the English one, and it links the Cursor, Cognition, VS Code and Replit sites alongside it. Nothing in the README grants any right to the underlying vendor prompts, so the disclaimer is doing real work rather than being boilerplate, and anyone planning to reuse the content in a product, a course or a training set should read DISCLAIMER.md and, if it matters, ask the vendors.

How a translation gets added

The contribution flow is the standard fork-and-pull-request cycle, and the README spells it out in five steps, which is worth quoting because the branch naming convention is specific. You fork the repository, then create your feature branch, commit your changes with a message describing the tool you added or the translation you improved, push the branch, and open a pull request. The commands are ordinary Git:

git checkout -b feature/amazing-translation
git push origin feature/amazing-translation

The commit message in the README is a template describing the addition of a translation for a given tool, and the branch name is feature/amazing-translation by convention. The project explicitly asks for more Chinese translations of AI tools and for improvements to existing ones, which means the contribution is not only about new directories but also about fixing a translation that has drifted from the upstream text. That second case is the more valuable one over time, and it is also the one the repository has no tooling to help with, since there is no commit that links a translated file to the upstream commit it came from.

Where this repository is the wrong tool

The limits are inherent to what the thing is. First, it is a mirror of somebody else's collection, so it inherits that collection's accuracy problems and adds a translation layer on top. The upstream project documents prompts for products that change frequently, and a translation of a document that was rewritten last week is not a translation of the current document. The second limit is that the repository gives you no way to tell versions apart. It publishes no GitHub releases, it carries a single 1.0.0 badge, and there is no mapping between a file in a directory and the upstream version it was translated from, so a reader who finds a prompt and wants to cite it or rely on it has no version to cite. The third is that the content is not behaviour. A system prompt tells you what an assistant was told to do, not what it does, and a translated prompt tells you even less about the actual product. The fourth is scope, and it is the least interesting but the most practical. If you want to build an agent, these documents are a reference and a source of ideas, not a specification you can implement against. And if you read English, the upstream project is the one to read, because the translation adds distance without adding anything except the disclaimer.

Editorial conclusion

Adopt this repository if you read Chinese faster than English and you want to study how commercial coding assistants are instructed, because the translation is more usable to you than the original even if it is imperfect. Do not treat it as an authoritative source, since the documents are translations of vendor internals that change without notice, and the repository publishes no releases to tell you which upstream version a text corresponds to. Before you rely on any document in it, compare the passage that matters against the original English project, and read DISCLAIMER.md, which states the project is for study and research only.

Frequently asked questions

Official sources

1. IsHexx/system-prompts-and-models-of-ai-tools-chinese on GitHub
2. Issues
3. License: MIT
4. README