AI-IDE-Agent, a Chinese prompt collection of sixty-one expert personas

What the collection actually contains

AI-IDE-Agent is described as a collection of prompts for Claude, Cursor and Trae, and the README states it contains sixty-one prompts for professional domains, each carefully designed and optimised for its specific field.

The framing that matters is in the README's own wording. Every prompt is written in Chinese, and the stated reason is that Chinese is easier for Chinese users to understand and use. That is a deliberate scope decision rather than an oversight, and it determines who the collection is for.

The usage model is copy and paste. The README's steps are: pick the relevant professional category folder, open the expert prompt file, copy the full contents, paste it into the conversation in Claude, Cursor or Trae, and then describe your specific requirement. From there the assistant is expected to answer as the configured expert.

The structure is visible in the repository. There is a README.md at the root and six category directories: programming language experts, cloud architecture and DevOps, data and AI, business and product, security and quality, and mobile and game development. There is also a separate directory for Chinese agents, alongside the six.

The statistics section puts the count precisely: sixty-one Markdown files plus one README, so sixty-two in total, across six main categories, in standard Markdown format with YAML front matter. The README lists the covered areas as programming, cloud services, data science, product design, security and mobile development, and the update log says the files have been translated into Chinese and re-sorted by professional domain.

The programming and cloud categories

The programming language category holds thirteen prompts and is the largest of the technical ones. Each file names a language and a specialty.

C#, aimed at .NET development with Unity, ASP.NET and async programming. C++ for systems work, with memory management, the STL and RAII. C for low-level work with pointers, memory and system calls. Elixir for functional programming with OTP, Phoenix and concurrency. Golang for cloud-native work with concurrency, microservices and performance.

Java covers enterprise development with Spring, concurrency and the JVM. JavaScript covers front-end work with ES6+, async and frameworks. PHP covers web development with Laravel and performance. Python covers data science with async, decorators and testing. Ruby covers Rails, metaprogramming and DSLs. Rust covers systems work with memory safety, concurrency and performance. Scala covers functional programming with Spark, Akka and type systems. TypeScript covers front-end work with types, generics and frameworks.

The cloud and DevOps category has five entries and a different character, since the roles are jobs rather than languages. The cloud architect covers multi-cloud design, microservices and scalability. The DevOps troubleshooting specialist covers monitoring, logging and performance tuning. The MLOps engineer covers model deployment, experiment tracking and automation. The Terraform expert covers modularity, state management and multiple environments. The deployment engineer covers CI/CD, containerisation and blue-green deployment.

That pairing of language and role prompts in one collection is the point. A developer can pick the Rust prompt for a systems question and the deployment engineer prompt for the same project's rollout, and the README's best practices section explicitly suggests combining several experts and using different ones at different stages of a project.

Data, business and security prompts

The data and AI category holds eight prompts. The AI engineer covers model integration, API design and optimisation. The data engineer covers ETL, data pipelines and quality assurance. The data scientist covers statistical modelling, visualisation and insight. The database administrator covers performance tuning, backup and recovery, and security. The database optimiser covers query optimisation, index design and monitoring. The machine learning engineer covers model training, feature engineering and deployment. The prompt engineer covers prompt design, evaluation and optimisation, which is a slightly meta entry given what the repository is. The search specialist covers algorithm work, relevance and performance.

The business and product category is the largest with twenty-four prompts, and it is where the collection strays furthest from engineering. It includes the API documentation engineer, the GraphQL architect, a Mermaid expert for diagrams, an SQL expert, a UI-UX designer covering accessibility and prototyping, a business analyst, a content marketer, customer support, a quantitative analyst covering financial modelling and risk, a risk manager, a legal adviser covering privacy policy and compliance, a sales automation specialist, a network engineer, a documentation architect, an architecture reviewer, a backend architect, a front-end developer, a performance engineer, a developer experience specialist, a tutorial engineer, a reference builder, a context manager, a legacy system modernisation specialist and a payment integration specialist.

The security and quality category has six: a security code reviewer covering quality, best practice and security, an error detective for localisation, root cause analysis and fixes, a test automation specialist covering CI/CD and coverage, a debugger for diagnostics and profiling, and an incident responder for emergency handling, recovery planning and improvement.

The mobile and game category closes the set with five: a Flutter expert, an iOS developer, a mobile developer, a Unity developer, and a Minecraft Bukkit specialist for plugin development.

The spread across roles that are not engineers, from the legal adviser to the content marketer, is a deliberate signal about who the author expects to use the collection.

The best practices section and how these are meant to work

The README's usage advice is worth reading because it describes how a prompt collection is supposed to be applied rather than just consumed.

The first point is to choose the right expert for the question, and to combine several for a fuller answer, using different ones at different stages for a complex project.

The second is to give detailed information: describe the specific context and requirements, provide the relevant code, error messages or data, and state the expected output format and requirements. This is the part that matters most in practice, because the quality of a persona prompt's output is bounded by what you put into it. A role prompt does not know your code.

The third is iteration: refine from the first result, feed back, and improve step by step until the answer fits.

The fourth is to save the useful results, building a personal knowledge base and sharing it with the team.

The contribution guidance is specific enough to be worth quoting. Fork the project, create a new expert prompt file, ensure content quality and professionalism, and submit a pull request. The requirements are that the content is written in Chinese, contains a complete description of the professional domain, provides practical methods and output formats, and follows the existing file naming conventions.

The support section asks for issues and pull requests, and the closing line asks for a star if the collection is useful. The repository is not archived and the last push was on 2026-04-10, with no GitHub releases. The repository's licence field says it is not one GitHub can classify, which is worth resolving before you copy prompts into a commercial workflow.

How this compares with writing your own system prompt

The alternative is obvious and worth naming carefully, because for most purposes it is the better one.

Writing your own role prompt takes ten minutes. You know your codebase, your team's conventions, what you want output to look like and which trade-offs you have already settled. The value of a generic expert prompt is that it starts from a shared vocabulary you would otherwise have to write, and that is a real saving when you are starting a conversation and do not yet know what you need.

The difference against the alternative of a prompt library maintained by a model vendor is mostly language. If you use the tools in English, the vendor's own prompt examples and documentation are already in your language, and the framing of a role is not what limits the output. A prompt collection in Chinese only helps if you want the assistant to think and answer in Chinese, which is a legitimate reason and not the same as a general advantage.

There is a more technical objection too, and it is the one that decides the question for an experienced user. A persona prompt tells the model what role to take and what vocabulary to use. Modern assistants are already instructed to be helpful and to apply domain knowledge. Adding a role description changes the framing of the answer, and often improves the shape of it, but it does not add knowledge and it does not add the ability to see your actual repository.

The case for these prompts is strongest in two situations. The first is a non-expert asking a domain question in Chinese, where the prompt supplies the vocabulary to ask it well. The second is an experienced developer who wants a consistent framing across a team, so that everyone asking about Terraform gets the same checklist. Both are legitimate, and the collection is a reasonable starting point for either, provided you treat the prompts as drafts rather than as finished artifacts.

Editorial conclusion

Adopt AI-IDE-Agent if you work in Chinese and want a starting system prompt for a role you write about often, and you are willing to rewrite it rather than paste it verbatim. Skip it if you work in English, since every prompt is written in Chinese, or if you want a maintained persona library with versioned updates. Verify first by pasting one prompt into the tool you use and running the same question with and without it, then keep it only if the difference is visible in the output.

Frequently asked questions

Official sources

1. bjlida/AI-IDE-Agent on GitHub
2. Issues
3. README