When using AI for programming, we often run into these “high blood pressure” moments:
• Ask AI to write a simple business feature, and it immediately produces a massive 800-line file;
• To get the new code running, AI quietly modifies your originally perfect unit tests (or even deletes them);
• The UI style was originally a serious Element admin-dashboard design, but AI insists on giving you an eye-searing cyberpunk neon gradient;
• The day after carefully teaching Cursor your best practices, you have to explain everything all over again when switching to Claude Code……
AI is highly capable, but without “enterprise-grade boundary constraints,” it can turn into a tireless “code destroyer.”
So I open-sourced cc-use-exp
GitHub: GitHub - doccker/cc-use-exp: ⛔ No longer maintained → dev-agent-kit/recipes 让 Claude Code、Antigravity、Gemini CLI、Codex、Cursor 开箱即用的分层配置模板，总结十多年的日常开发经验 · GitHub (Feel free to click 
Star to show your support!)
cc-use-exp is not simply a “prompt collection,” but a comprehensive AI
coding assistant configuration system and architecture governance tool. It is designed to put a “tight constraint” on AI, ensuring that its output consistently maintains high quality, high cohesion, and a unified style.
Core capabilities and highlights:
1. A unified foundation for 6 major AI tools
One architecture, seamlessly compatible with and distributable to the six most powerful AI coding tools currently available: Claude Code | Cursor | GitHub Copilot | Gemini CLI | Antigravity CLI | Codex
2. Rock-solid “Defensive Rules”
It includes an extremely strict set of global defensive rules. Once loaded, AI must follow them rigorously:
• Absolute code line limits: Java files must not exceed 300 lines, and Vue/TSX files must not exceed 200 lines. Anything longer must first be broken down into a skeleton before code is written.
• Strictly prohibit modifying tests or downgrading: Using @ts-nocheck to bypass errors and removing existing assertions are forbidden.
• Reply exclusively in Chinese: Completely eliminate the awkwardness of AI suddenly switching to English when working across projects.
• UI aesthetic isolation: Enforce a restrained enterprise-grade style, decisively curbing AI’s impulse to overuse “glassmorphism” and “glowing outlines.”
3. Skill stack: Write Once, Sync Everywhere
Tired of switching back and forth between multiple tools? This project provides the powerful sync-skill.sh synchronization script.
You only need to write a Skill once in the Claude source directory (for example, api-proxy-safety, an API security specification). The script automatically performs AST parsing, frontmatter trimming, and platform terminology translation, adapting and distributing the skill to Cursor, Codex, Gemini, and all other platforms with a single click.
4. A pioneering collision-avoidance approach for “multi-AI concurrent collaboration”
When Claude handles the main architecture, Gemini handles auditing, and Codex handles quick fixes, how can you prevent them from conflicting? The project includes multi-agent collaboration rules, including timestamp-based conflict prevention and unified task progress traceability standards.
How do I use it?
It works out of the box. Run a single command to fully deploy the powerful rule library to your local AI toolchain:
git clone https://github.com/doccker/cc-use-exp.git
cd cc-use-exp
./tools/sync-config.sh
Contributing
Building the perfect AI programming workflow is an ongoing process. If you also have your own tried-and-tested AI training rules (such as specific backend security standards or powerful frontend component extraction techniques), or have new ideas about collaboration across AI tools, you are very welcome to open an Issue or submit a PR!
If you feel this project has helped your everyday AI coding even a little, please give cc-use-exp
GitHub - doccker/cc-use-exp: ⛔ No longer maintained → dev-agent-kit/recipes 让 Claude Code、Antigravity、Gemini CLI、Codex、Cursor 开箱即用的分层配置模板，总结十多年的日常开发经验 · GitHub a precious Star! You are also welcome to share the AI coding scenarios that currently cause you the most pain in the comments.
