loaditout.ai
SkillsPacksTrendingLeaderboardAPI DocsBlogSubmitRequestsCompareAgentsXPrivacyDisclaimer
{}loaditout.ai
Skills & MCPPacksBlog

AI-Plugin-Gatekeeper

MCP Tool

wangxiaotiantian/AI-Plugin-Gatekeeper

🛡️ Stop AI agents from burning tokens. AI 插件调用逻辑门控协议。

Install

$ npx loaditout add wangxiaotiantian/AI-Plugin-Gatekeeper

Platform-specific configuration:

.claude/settings.json
{
  "mcpServers": {
    "AI-Plugin-Gatekeeper": {
      "command": "npx",
      "args": [
        "-y",
        "AI-Plugin-Gatekeeper"
      ]
    }
  }
}

Add the config above to .claude/settings.json under the mcpServers key.

About

🛡️ Plugin Gatekeeper | 插件守卫 (Dual-Language)

Stop AI from burning tokens and over-analyzing. 防止 AI 插件乱烧 Token 和过度分析。

---

💡 Why use this? / 为什么要用它?

AI plugins (like Superpowers) are often "too diligent"—re-reading the entire project for a simple fix or re-planning everything after a short break. This Skill gives the control back to you.

AI 插件往往“过度勤快”:修个小错别字也要遍历全项目,或者中断回来就要重读所有文档。本协议通过逻辑门控,把插件的“开关”重新握在用户手中。

---

✨ Key Features / 核心功能
1. Manual Authorization / 手动授权拦截

AI must ask for permission before performing heavy scans or global searches. AI 执行重型扫描或全局搜索前必须先“请示”。

2. Zero-Latency Resume / 断点无感续传

Skip re-reading docs by using a 3-line [Session_Anchor]. 通过 3 行“会话锚点”跳过文档重读,直接进入开发/任务状态。

3. Precision Mode / 精准模式

Force AI to focus only on the reported error or the current file. 强制 AI 仅关注报错区域或当前文件,严禁“想太多”。

---

🚀 Quick Start / 快速开始
  1. Copy (复制): Copy the content of config/gatekeeper.json.
  2. Inject (注入): Paste it to your AI (Claude, Cursor, GPT-4) and say:

> "Load this skill and follow the logic gate rules. / 加载此技能并遵守逻辑门控规则。"

  1. Control (控制):
  • Choose Local-Only to save tokens. (选“本地处理”来节省 Token)
  • Say "Archive / 存档" before you leave. (离开前说“存档”)
  • Paste the anchor and say "Continue / 继续" when you are back. (回来后贴上锚点说“继续”)

---

📂 Repository Structure / 仓库结构
  • /config/gatekeeper.json: Machine-readable logic. (机器可读逻辑)
  • /docs/guide_en.md: Detailed English guide.
  • /docs/guide_zh.md: 详细中文指南。

---

📄 License

MIT License. Created for the Vibecoding community.

备注:我是因为修复 Bug 和断点重连太费 Token 才想做这个,实战中可以这样玩: 当 AI 弹出那句“⚠️ [Gatekeeper] 检测到重型任务...”时: 如果你很确定哪里错了,直接回:“3”(即本地处理,不许它搜索)。 如果你毫无头绪,回:“1”(让它搜一次)。 当你想下班时: 输入:“Archive / 存档”。 AI 会吐出 3 行文字(比如:已完成登录逻辑、正在调接口、下一步处理报错)。 重点:你自己把这 3 行存到记事本。 第二天回来时: 把那 3 行贴给 AI,说:“Resume / 继续”。 你会发现它不再废话,也不会去读 README,直接开始敲代码。

Note: I’m doing this because fixing bugs and re-establishing breakpoints wastes too many tokens.This trick works in actual use: When t

Tags

ai-agentsmcpprompt-engineeringsuperpowerstoken-savingvibecoding

Reviews

Loading reviews...

Quality Signals

0
Installs
Last updated5 days ago
Security: BREADME
New

Safety

Risk Levelmedium
Data Access
read
Network Accessnone

Details

Sourcegithub-crawl
Last commit4/8/2026
View on GitHub→

Embed Badge

[![Loaditout](https://loaditout.ai/api/badge/wangxiaotiantian/AI-Plugin-Gatekeeper)](https://loaditout.ai/skills/wangxiaotiantian/AI-Plugin-Gatekeeper)