How to use vibe-coding-cn: planning-first AI coding guide

What vibe-coding-cn is and what it is not

The README describes Vibe Coding as a workflow for pairing with AI to turn ideas into working software. It stresses planning and modularity so the AI does not drive the project into disorder.

The repository is guidance, not software. Nothing in the README describes an application you install and run, and no prerequisites are listed.

Get the repository and read it online

The source gives one install command: clone the repository. You can also read the material on GitHub without cloning. The main README is in Chinese, and the English version is linked from the language badges at ./i18n/en/README.md.

• Clone the repository to keep a local copy of the documents and prompts.

Start with the core principle: plan before you code

The central idea is that planning comes first. The README warns against letting AI plan freely, because the codebase can become unmanageable. It lists principles such as building the structure before the code, keeping modules separate, and treating documentation as context rather than an afterthought.

• Write a one-sentence goal and a non-goal before starting.
• Plan the framework first to avoid later technical debt.
• Split code by responsibility and change one module at a time.
• Define interfaces before implementation.
• Write documentation as context for the AI, not after the fact.

Give the AI clear context and debug with precise reports

The README says context is the first-order element of vibe coding, so poor input gives poor output. For debugging, it recommends giving the expected result, the actual result and a minimal reproduction. It also advises stating clearly what the AI may change and what it may not.

The guide suggests letting AI write tests while a human reviews the assertions, and starting a new session when code grows large.

• State what can and cannot be changed.
• For bugs, give expected vs actual behaviour plus a minimal reproduction.
• Let AI draft tests; review the assertions yourself.
• Split into a new chat session when the code becomes large.

Use the recursive prompt-optimisation method (optional)

The README describes a meta-method with two master prompts: an α-prompt that generates prompts or skills, and an Ω-prompt that optimises them. The lifecycle is bootstrap v1 versions with AI, use the Ω-prompt to improve the α-prompt, generate targets with the improved α-prompt, then feed outputs back into the loop.

The author notes that the experience shared is not universally applicable, so adopt ideas selectively and test them in your own context.

Choose tools from the recommendation lists

The README lists IDEs, terminals, editors and AI models that the author recommends. Examples include Visual Studio Code, Cursor, Warp, Neovim with LazyVim, Claude Opus 4.5 and Codex CLI. These are the author's picks, and the README does not provide independent benchmarks to compare them.

Check each tool's own site and terms before adopting it.

• Visual Studio Code: suited to reading code and manual edits.
• Cursor: listed as a well-known AI IDE.
• Warp: a terminal with AI features for command-line work and debugging.
• Neovim with LazyVim: a keyboard-driven editor setup.