A field manual for vibe coding that mistrusts its own vibes

What it does

vibe-coding-cn is a Chinese tutorial and knowledge base for AI pair programming—what the README calls “vibe coding.” It maps a full workflow from idea to product through a layered framework of Prompts, reusable Skills, Context management, Quality Gates, and Git-based engineering loops. The repository is essentially a structured field manual: docs, reference architectures, prompt libraries, and skill templates meant to keep AI-generated code from collapsing into unmaintainable glue.

The interesting bit

The repo’s “Five Core Propositions” read like a risk assessment dressed as a tutorial. It argues that LLMs will eventually swallow the prompt-engineering scaffolding we use today, so the durable skill is orchestration—reverse-searching mature libraries and writing only the glue code and business rules needed to bind them. It also enforces “isolation review,” insisting that AI output must be verified in a separate context because an LLM judging its own work is a conflict of interest, not a quality gate.

Key highlights

• A complete Chinese-language learning path from CLI setup to full workflow execution, organized under a “Dao Fa Shu Qi” (道法术器) framework.
• Curated prompt libraries, reusable Skill modules, and a “keyword system” for concept navigation.
• Explicit governance rules: AI output is a “candidate solution” requiring independent review, testing, and schema validation before it counts as fact.
• A “glue coding” methodology that prioritizes reverse-searching existing toolchains over prompting AI to reinvent them.
• Reference templates for single-machine enterprise architecture and modern stack selection.

Caveats

• The README is dense with badge links and conceptual jargon (“Dao Fa Shu Qi,” “模型吞噬”) that may scan as philosophy-heavy for developers who just want a quick Cursor cheat sheet.
• Despite the Python language tag, the visible content is overwhelmingly documentation and methodology; there is little evidence of runnable tools in the excerpt provided.
• The “Five Propositions” explicitly warn that prompt tricks and workflow scaffolding may become obsolete as models improve, which raises the question of how durable the specific templates are.

Verdict

Worth bookmarking if you are a Chinese-speaking developer or team trying to institutionalize AI pair programming without letting your codebase drift into hallucinated spaghetti. Skip it if you are looking for a drop-in CLI tool or a language-agnostic quickstart; this is a reading-heavy methodology repo, not a utility.