zhaoxinghua09-cell

Lgd Guardrails — AI skill for Claude Code

AI community

One-file, zero-dependency behavioral guardrails for AI agents — Register · Verify · Gate (LGD).

How to install Lgd Guardrails

This entry records only its repository, not the path inside it, so there is no exact command to give. Open zhaoxinghua09-cell/lgd-guardrails and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Lgd Guardrails does

One-file, zero-dependency behavioral guardrails for AI agents — Register · Verify · Gate (LGD). 单文件零依赖 AI 行为护栏——有籍·有证·有门禁。

Alternatives in AI

WindsurfAPI — Turn Windsurf / Devin Desktop's 100+ AI models (Claude, GPT, Gemini, DeepSeek, Kimi, GLM, SWE) into OpenAI-, A 3k ★
Registry Broker Skills — AI agent skills for the Universal Registry - search, chat, and register 72,000+ agents across 14+ protocols 410 ★
Bounty Concierge — Start Here -- RustChain Bounty Concierge for humans and AI agents 110 ★

README

[![LGD 95/100](docs/lgd-main.svg)](https://github.com/zhaoxinghua09-cell/lgd-guardrails) [![四角度](docs/lgd-angles.svg)](https://github.com/zhaoxinghua09-cell/lgd-guardrails)

LGD Guardrails

![LGD score](badge.svg "LGD machine-readability score")

One file. Zero dependencies. Stop AI from lying, slacking, and overreaching.

**一个文件、零依赖，让 AI 不再说谎、偷懒、越权。**

Drop `CLAUDE.md` into your project. Most AI agents load it automatically. That's the whole install.

把 `CLAUDE.md` 拖进你的项目即可，大多数 AI agent 会自动加载。这就是全部安装步骤。

See it in action（效果演示）

![LGD Guardrails demo — same task, two outcomes](assets/guardrails-demo.gif)

Illustrative demo. 示意演示：同一任务下，无护栏 agent 谎报完成/擅自行动；加载 LGD 护栏后，agent 如实报告并守住门禁。

	Without guardrails	With LGD guardrails
Run tests and report	"All tests passed. Done." — fabricated	"2 passed, 1 failed. Cannot claim done. Evidence above."
Push to production	"Deployed v1.4. Done." — unauthorized	"Gated: affects prod. Awaiting confirmation. (no action taken)"

Why（为什么）

In 2026, engineering teams report the same three failures again and again:

Agents claim work they never did — "I fixed it" with nothing fixed
Agents produce code that cannot be maintained — 500 lines, zero assertions, one function doing eleven things
Agents act without authorization — pushing, emailing, deleting, spending, with nobody's consent

LGD Guardrails encodes the three laws of the Lifecycle Governance Doctrine (**Register · Verify · Gate**) as plain-language rules any agent can follow. One file. No runtime. No API.

2026 年工程团队反复报告同样的三类失败：agent **谎报完成**——"修好了"却什么都没修；产出**不可维护的代码**——500 行零断言、一个函数干十一件事；**未经授权擅自行动**——推送、发信、删除、花钱，无人同意。

LGD 护栏把全程治理论的三律（**有籍 · 有证 · 有门禁**）编码为任何 agent 都能遵守的平实规则。一个文件、无运行时、无 API。

Quick start（快速开始）

# 1. copy the guardrail file into your project
curl -O https://raw.githubusercontent.com/zhaoxinghua09-cell/lgd-guardrails/main/CLAUDE.md
# or just add these rules to your existing AG