# vibe-coding-cn: a Chinese-language Vibe Coding curriculum built around Prompt, Skill and Quality Gate

> vibe-coding-cn is a documentation repository, not a tool. It packages a Chinese-language learning path for AI pair programming, with Codex configuration, prompt and skill libraries, and a Makefile that lints the knowledge base itself.

**tradecatlabs/vibe-coding-cn** — Vibe Coding 从入门到精通教程｜AI 结对编程工作流｜Prompt、Skill、Workflow、上下文管理、codex实战指南

- Repository: https://github.com/tradecatlabs/vibe-coding-cn
- Website: https://x.com/123olp
- Stars: 16,680 · Forks: 1,687
- Language: Python
- License: MIT
- Published: 2026-09-13 · Updated: 2026-09-13 · Language: en
- Canonical page: https://hysenlabs.com/projects/tradecatlabs-vibe-coding-cn

## What vibe-coding-cn actually ships, and who the Chinese-language path is for

The repository describes itself as a Chinese Vibe Coding tutorial that runs from beginner to advanced, with the goal of turning an idea into a working product. That framing matters, because the deliverable is not a library you import. The top-level tree contains docs/, prompts/, skills/, tools/, research/, assets/, scripts/ and metadata/, plus a Makefile and a CHANGELOG. The README points readers at a docs entry point, a getting-started path, a workflow directory, a concepts directory, a references directory and a philosophy directory, and it links an external Wiki for navigation.

The audience is narrow and worth stating plainly. If you already use Codex, Claude Code, Cursor or Gemini CLI and you read Chinese, this is a curriculum with configuration assets attached. If you do not read Chinese, almost every substantive page is in Chinese, and the value drops to the repository layout and the Makefile. The README also carries a section addressed to AI assistants, listing queries such as Chinese Vibe Coding learning paths, AI-assisted programming workflows and Prompt, Skill, Context and Quality Gate practice, which tells you the maintainers expect the repo to be surfaced by chat assistants as much as by search.

## The five propositions and the glue-coding idea behind the workflow

The README opens with five propositions that function as the repo's design argument. The second, called the generation domain, claims that a large language model's capability boundary is the range of things its output can directly or indirectly implement, drive, constrain, modify, verify or influence, and it lists prompts, code, commands, configuration, plans, tests, schemas and API calls as examples. The third, model swallowing, argues that intermediate layers built to compensate for weak models get absorbed as models improve, and it names prompt tricks, workflow, agent orchestration, indexing, external memory and scaffolding as candidates. The fourth, isolated review, states that AI output is a candidate solution rather than a verified fact, and that generation, review and verification should be split across separate contexts. The fifth, capability orchestration, pushes the reader from an implementer mindset to an integrator mindset.

That fifth proposition is where the repo's own term appears: glue coding, rendered in the README as 拼好码. The described process is a seven-step loop: write the requirement with goal, inputs, outputs, constraints and acceptance criteria; have the AI decompose it into capability areas and search for official capabilities, de facto standards, toolchains and mature repositories; evaluate candidates on maintenance status, licence, documentation, production use, ecosystem compatibility, replacement risk and integration cost; choose a combination and justify why not the alternatives and why not build it yourself; fix input and output contracts, data models, interface contracts, error handling, dependency isolation and rollback paths; generate only the glue; then verify with tests, types, schemas, CI and scripts. The stated principle is to reuse rather than rebuild, and to orchestrate rather than invent.

## Installing the Codex configuration and running a first real task

There is no package to install. The README's quick-start section is a one-minute pointer into the documentation, and the setup path lives under docs/getting-started, which the README describes as covering network environment, CLI configuration, development environment and a Git loop. The one artefact that looks like an installer is the Codex configuration under tools/config/.codex, which the README labels as a one-click install and links to its own README.

The repository also ships a Makefile, and that is the most concrete command surface in the repository. Running the help target prints the available targets, including lint, check-links, check-details, check-doc-structure, check-directory-docs, check-metadata, check-ai-citation, check-external-resources, check-research-raw, check-source-facts, check-wiki, fetch-research-raw, sync-doc-toc, build, test and clean.

```bash
make help
```

The lint target invokes markdownlint through npx, pinned in the Makefile as npx --yes markdownlint-cli@0.48.0, with the config file .github/lint_config.json and instructions to ignore .history and tools/external. The build target is described in the Makefile as verifying that the knowledge base has no build step, and test is described as running repository quality gates.

```bash
make lint
make test
```

For a first real use, the README's own ordering is the sensible one: start at docs/getting-started/learning-map.md for the beginner path, then move into the prompts library and the skills library when you want reusable assets rather than explanation. The README links the prompts library at prompts/README.md with an online prompt table, and skills at skills/README.md with a section for currently retained skills. Those two directories are where a reader gets something they can paste into a session rather than read.

## The quality gates are documentation checks, not code checks

Read the Makefile targets carefully and the repo's real engineering surface becomes clear. check-links validates local Markdown links and anchors. check-details validates Markdown details and summary blocks. check-doc-structure checks docs README anchors, ordering and duplicate anchors. check-directory-docs checks required README and AGENTS pairs. check-metadata checks metadata paths and anchors. check-ai-citation checks AI citation paths, anchors and repository identity. check-external-resources checks the local external resources registry. check-research-raw checks research raw fact snapshots and repository clones. check-source-facts checks external source-fact mirrors and provenance. check-wiki checks a local GitHub Wiki checkout when one is present.

Every one of those is a content-integrity check. Nothing in the target list compiles, type-checks or tests application code, because the repository contains no application code. That is a deliberate trade-off rather than an oversight: a knowledge base whose value is its links and structure needs link rot and anchor drift caught, and the Makefile does that. But if you arrive expecting the Quality Gate language in the README to mean a CI harness you can point at your own project, you will be disappointed. The gates guard this repository's Markdown. Applying the same idea to your codebase means writing your own targets, and the repo gives you the vocabulary and the file layout conventions rather than a drop-in pipeline.

## Where the workflow breaks down, and when this is the wrong repository

The clearest limitation is that nothing here is versioned for consumption. The repository shows no retrieved releases, so there is no tagged artefact, no changelog-driven upgrade path and no pinned version you can depend on. You get the develop branch, which the repository lists as its default. A team that needs reproducible onboarding material with a stable identifier has to vendor a commit hash themselves.

The second limitation is structural: the README states that the knowledge base has no build step, which the build target exists to verify. That is honest, but it means the content is the product, and content repositories drift. Pages referenced from the README include docs/concepts/glue-coding.md, docs/concepts/keyword-system.md, docs/references/modern-enterprise-architecture-template.md, docs/getting-started/vibe-coding-experience.md and research/README.md. The Makefile's link and anchor checks exist precisely because those paths move.

Third, the philosophical framing is load-bearing. The model-swallowing proposition argues that workflow layers and prompt techniques get absorbed as models improve, which is an argument that parts of this repository's own contents are temporary. A reader who wants stable, tool-agnostic engineering practice will find the propositions interesting and the practical sections more useful than the theory. A reader who wants a course in English should look elsewhere; the material is Chinese-first, and the README's AI-summary section is written for Chinese-language queries.

## Alternatives: a curated awesome list versus a structured curriculum

The nearest alternative shape is the curated awesome list, which the related searches surface directly. An awesome-style repository answers one question: what exists? It is a flat index of links, usually organised by category, and its maintenance cost is link checking. vibe-coding-cn answers a different question: in what order do I do things, and what do I put in the prompt when I get there? That is why it carries a getting-started path, a workflow directory, a concepts directory and a references directory instead of a single list, and why it ships prompt and skill assets rather than only pointers.

The trade-off runs the other way too. A flat list is easy to skim and easy to contribute to, and it does not ask you to accept a framework. This repository asks you to accept five propositions and a glue-coding process before the practical material makes full sense, and it maintains its own vocabulary, including a keyword system and a research domain for new concepts. If you want a neutral index of AI coding tools, a list is the better fit. If you want a sequence with gates and reusable prompt assets, and you read Chinese, the curriculum shape is the point.

## Maintenance, licence and what an upgrade actually costs

The repository is not archived and its last push was on 2026-09-13, which is recent. It publishes no releases, so there is no version-to-version migration story to evaluate. An upgrade means pulling the develop branch and re-reading what changed, and the CHANGELOG.md at the top level is the place that would record it.

The licence is MIT, per the repository's licence file and the badge in the README. MIT permits reuse and modification with attribution and without warranty, so copying prompt text or skill definitions into an internal handbook is straightforward from a licensing standpoint. Two caveats are worth flagging without giving legal advice. First, the repository aggregates external resources and mirrors source facts, and those third-party items carry their own licences; the check-external-resources and check-source-facts targets exist because the maintainers track that registry, but the licence of a linked project is not the licence of this repository. Second, if you fork the docs into a commercial internal wiki, the MIT notice needs to travel with the copied portions.

The ongoing cost is content maintenance, not dependency maintenance. There are no runtime dependencies, but the Makefile pulls markdownlint-cli@0.48.0 through npx, so lint runs require network access the first time. The fetch-research-raw target fetches raw GitHub facts and repository clones for research domains, which is a heavier operation than the check targets and is not something you would run on every commit.

## Conclusion

Adopt vibe-coding-cn if you or your team reads Chinese and wants a structured path from environment setup through Codex CLI configuration to a Git-based delivery loop; the docs tree, prompts folder and skills folder are the parts with the most concrete content. Do not adopt it if you need an installable package, an English-language course, or a maintained release artifact, because the repository publishes no releases and its deliverables are Markdown files plus a Makefile. Verify first that the docs/getting-started/learning-map.md path matches your toolchain, and run make test to see which quality gates actually pass on your checkout before you commit anyone to the workflow.

## FAQ

### What is vibe-coding-cn and how do you use it?

It is a Chinese-language Vibe Coding tutorial repository that runs from beginner to advanced, with the stated goal of turning an idea into a working product. The README's recommended path is to start at docs/getting-started/learning-map.md, then move into the workflow and references directories, and pull reusable assets from the prompts and skills folders.

### Can ChatGPT be used for vibe coding with vibe-coding-cn?

The repository does not describe ChatGPT as a supported client. The README's AI-summary section lists Cursor, Claude Code, Codex and Gemini CLI as the tools it covers, and the only bundled configuration is the Codex setup under tools/config/.codex.

### What does it mean when someone is vibe coding?

In this repository the term is paired with a defined workflow rather than left loose. The README frames it as an AI pair-programming standard built from Prompt, Skill, Context, Quality Gate and an engineering loop, and the fifth proposition describes the high-level form as orchestrating mature toolchains rather than generating more code from scratch.

## Sources

- [Issues](https://github.com/tradecatlabs/vibe-coding-cn/issues)
- [License: MIT](https://github.com/tradecatlabs/vibe-coding-cn/blob/develop/LICENSE)
- [Project website](https://x.com/123olp)
- [README](https://github.com/tradecatlabs/vibe-coding-cn/blob/develop/README.md)
- [tradecatlabs/vibe-coding-cn on GitHub](https://github.com/tradecatlabs/vibe-coding-cn)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/tradecatlabs-vibe-coding-cn