admin/Custom-Engineered-Agents

Fork 0

Files

Gemini AI 2df732531f feat: add GLM MCPs Enhanced agent with credits to Tycen

2025-12-23 19:01:36 +04:00

27 KiB

Raw Permalink Blame History

🤖 Ultimate Custom Engineered AI Agents & More...

Uncapped tools Collection for Vibe Coders

🚀 Quick Start • 🎯 Agents • 📰 AI Digest • 💡 SOLO + GLM • 🛠️ IDEs • ☁️ Cloud VM • 🌟 MCPs • 💡 Hacks • 🕹️ Games

📰 AI Digest: GLM-4.7 vs Claude 4.5 Opus & Sonnet

🚀 GLM-4.7: The New SOTA for Agentic Workflows

The latest GLM-4.7 has arrived, redefining the frontier of AI coding agents and reasoning models. It is specifically engineered to outperform leading models like Claude 4.5 Sonnet and Claude 4.5 Opus in multi-step developer workflows.

⚔️ The Frontier Battle: Head-to-Head Benchmarks (2025)

Category	Benchmark	GLM-4.7	Claude 4.5	DeepSeek 3.2	Gemini 3 Pro	Kimi	Codex 5.2	Winner
Math	AIME 25	95.7	88.5	92.4	90.2	94.0	85.0	🥇 GLM-4.7
Coding	LiveCode	84.9	61.2	78.5	72.0	68.0	65.0	🥇 GLM-4.7
Science	GPQA	85.7	84.1	82.5	83.0	81.0	79.0	🥇 GLM-4.7
Logic	HLE	42.8	22.5	35.0	30.0	28.0	25.0	🥇 GLM-4.7
API Cost	Price / 1M	$0.60	$15.00	$0.35	$1.25	$1.00	$2.00	🥇 GLM-4.7

💡 Why GLM-4.7 is the Choice for Vibe Coders:

Crushing the Competition: Outperforms Gemini 3 Pro, DeepSeek 3.2, and Claude 4.5 in core reasoning and coding benchmarks.
Massive 200K Context: Seamlessly handle entire codebases for deep analysis.
Deep Thinking Mode: Forced systematic reasoning for high-complexity architectural tasks.
Extreme Value: 25X cheaper than Claude 4.5 Opus with significantly higher performance.
Real-time Tool Streaming: Optimized for TRAE SOLO, Cline, and Roo Code agents.

🛠️ Quick Tutorial: Configure GLM-4.7 in TRAE SOLO

🚀 Quick Start

Action	Link
🎟️ Get 10% Discount on GLM Models	Discount invitation code
🎟️ TRAE with GLM Explained	GLM Model with SOLO and TRAE
📄 GLM-4.6 Integration Guide (PDF)	Download PDF
🌐 Explore TRAE Plans	Start for Free

🎯 Latest Hackathon Participation

OffLogic Game is the ultimate reflex challenge for developers and UI designers. An interactive experience that tests your coding speed, precision, and aesthetic preference.

📂 Source Code of OffLogic

🤖 Available Agents

⚠️ Smart Disclaimer: While these specialized agents are engineered for optimized performance, they cannot guarantee 100% success in every coding scenario. Always ensure you have a valid backup or git commit before proceeding with any agent operations. We strongly recommend experimenting with each agent in a controlled setting to understand its specific capabilities before applying it to critical projects.

OPUS Series
_{Development patterns of Claude Opus 4.5 (Variation 1)} ⚡ Install Agent	_{Development patterns of Claude Opus 4.5 (Variation 2)} ⚡ Install Agent
_{Enhanced QA process based on Opus model} ⚡ Install Agent	_{Opus 4.5 engineering patterns - focusing on production-grade code, systematic thinking, and clear communication.} ⚡ Install Agent
Specialized Agents
_{CBT Rule Enforcer: Strict protocol for safe, auditable, and structured coding governance.} ⚡ Install Agent	_{Autonomous Selector: Discover and select tools for end-to-end task execution without human interaction.} ⚡ Install Agent
_{Amazon Kiro agent concepts} ⚡ Install Agent	_{GLM MCPs Enhanced: Design to handle Z.AI MCP services (Vision) for GLM models. Credit: Tycen (TRAE Discord) for the base agent.} ⚡ Install Agent
_{Transform web apps into PWAs} ⚡ Install Agent	_{Break infinite reasoning loops} ⚡ Install Agent
_{Overclock for lower-end models} ⚡ Install Agent	_{Research-driven architect for AI VIBE coding & complex solutions} ⚡ Install Agent
_{🚀 System Upgrade: Flash-to-Pro Cognitive Bridge Overclock lightweight models to simulate Pro-tier reasoning depth.} ⚡ Install Agent
Defense Series
_{Minimize AI "forgetting" to complete tasks} ⚡ Install Agent	_{Improved version to combat AI amnesia} ⚡ Install Agent
_{Attempt allowing GLM 4.7 model operate on an existing code of other models without breaking it. Essential for non-greenfield projects where code preservation is key.} ⚡ Install Agent

💡 Boost Efficiency with SOLO & GLM 4.7

🎟️ Get 10% Discount on GLM Models | Discount invitation code

Unlock the full potential of your development workflow by combining TRAE's SOLO Agent with the GLM 4.7 model.

Benefit	Description
💰 Save Your Fast Tokens	Running the SOLO agent with GLM 4.7 does not consume your TRAE paid fast tokens.
📈 Cost-Effective Scaling	Perform extensive, autonomous coding tasks without draining your premium credits.
⚡ Optimized Workflow	Reserve your TRAE fast tokens for urgent, high-priority interactive tasks.

🛠️ GLM 4.7 Alternative IDE Options

The GLM 4.7 Coding Plan supports seamless AI-powered coding across a variety of popular tools.

Supported Clients & IDEs

IDE	Link
Claude Code	Documentation
Cline	GitHub
OpenCode	GitHub
Roo Code	GitHub
Kilo Code	Website
Crush	GitHub
Goose	GitHub

⚙️ Setup Instructions

Subscribe: Get an active subscription to the GLM 4.7 Coding Plan.
Automatic Availability: In most supported tools, GLM-4.7 becomes available automatically upon subscription.
Claude Code Configuration: GLM-4.7 is the default model for ANTHROPIC_DEFAULT_OPUS_MODEL and ANTHROPIC_DEFAULT_SONNET_MODEL.

📚 For detailed documentation, visit the Z.AI Developer Docs.

☁️ Virtual Machine for AI Coding FREE - Anywhere

Full Stack Workstation is a free AI tool designed to transform your coding experience, suitable for small tasks.

Feature	Description
🌐 Open Source GLM Models	Powered by cutting-edge models, tailored for English and Chinese users.
🧠 Advanced Reasoning	Supports complex reasoning and deep research tasks.
💸 Zero Cost	A completely free, open-source alternative to paid AI assistants.
📱 Code Anywhere	Code from anywhere, on any device, with zero setup.

⚠️ Note: Best for small edits. Functionality is limited to one chat context window.

🌟 Awesome MCPs

Vision MCP Server

Visual Intelligence for Your IDE

The Vision MCP Server brings GLM-4.5V's advanced visual capabilities directly into MCP-compatible clients like Claude Code and Cline.

Feature	Description
🖼️ Image Analysis	Analyze and interpret various image formats.
🎬 Video Understanding	Gain insights from local and remote videos.
🔌 Seamless Integration	Easy setup with MCP-compatible tools.

Learn more & Install

💡 Developer Hacks

Quick Links

Hack	Description
📱 PWA Recipe	Make your app installable on Android
🤖 AI Push Prompt	Automate PWA integration with AI
🚀 Deploy to Vercel	Get a live URL in seconds
🧠 Claude Code + GLM	Use Claude Code with affordable GLM models

📱 Make your web app installable on Android devices. This step-by-step guide shows you how to configure Vite for PWA, create an install button component, and handle iOS fallback instructions.

<EFBFBD>️ CBT Rule Enforcer: Premium Protocol Explained

The CBT Rule Enforcer transforms your AI from an assistant into a Disciplined Steward. It enforces a non-negotiable governance framework for mission-critical production environments.

⚡ Core Behavioral Shifts

React (Halt & Discover): Forces immediate pause. No guessing. Initiates discovery, intent extraction, and explicit paraphrasing before a single line of code is written.
Enhance (Evidence-First): Mandates snippets_before and snippets_after with proof. Generates rollback_plan.json and enforces least_privilege safety.
Operate (Orchestration): Strict state machine workflow (Discovery -> Plan -> UserConfirm -> Implement). Every decision is logged with rationale and risk IDs.
Respond (Audit Trail): Communication shifts to structured artifacts. Requires specific user consent tokens for execution. Maintains an immutable audit_and_change_log.
Change (Stewardship): Prioritizes governance over speed. Transparent reasoning via context_map.json and risk_list.json. Rigidly prevents silent breaking changes.

<EFBFBD>📱 PWA Recipe

Make your app installable on Android (Complete PWA Recipe)

To enable the "Install App" feature, you need to configure Vite for PWA and create a custom install button component.

0. Install Dependencies

npm install -D vite-plugin-pwa

1. Configure vite.config.ts

import { defineConfig } from 'vite'
import react from '@vitejs/plugin-react'
import { VitePWA } from 'vite-plugin-pwa'

export default defineConfig({
  plugins: [
    react(),
    VitePWA({
      registerType: 'autoUpdate',
      includeAssets: ['favicon.ico', 'apple-touch-icon.png', 'masked-icon.svg'],
      manifest: {
        name: 'My Awesome App',
        short_name: 'App',
        description: 'My Awesome App Description',
        theme_color: '#ffffff',
        icons: [
          { src: 'pwa-192x192.png', sizes: '192x192', type: 'image/png' },
          { src: 'pwa-512x512.png', sizes: '512x512', type: 'image/png' }
        ]
      }
    })
  ]
})

2. Create src/components/InstallPWA.tsx

import { useEffect, useState } from 'react';

interface BeforeInstallPromptEvent extends Event {
  prompt: () => Promise<void>;
  userChoice: Promise<{ outcome: 'accepted' | 'dismissed' }>;
}

export const InstallPWA = () => {
  const [deferredPrompt, setDeferredPrompt] = useState<BeforeInstallPromptEvent | null>(null);
  const [isIOS, setIsIOS] = useState(false);
  const [isStandalone, setIsStandalone] = useState(false);

  useEffect(() => {
    if (window.matchMedia('(display-mode: standalone)').matches) {
      setIsStandalone(true);
    }
    const userAgent = window.navigator.userAgent.toLowerCase();
    setIsIOS(/iphone|ipad|ipod/.test(userAgent));
    const handler = (e: Event) => {
      e.preventDefault();
      setDeferredPrompt(e as BeforeInstallPromptEvent);
    };
    window.addEventListener('beforeinstallprompt', handler);
    return () => window.removeEventListener('beforeinstallprompt', handler);
  }, []);

  const handleInstallClick = async () => {
    if (!deferredPrompt) return;
    await deferredPrompt.prompt();
    const { outcome } = await deferredPrompt.userChoice;
    if (outcome === 'accepted') {
      setDeferredPrompt(null);
    }
  };

  if (isStandalone) return null;

  return (
    <div className="fixed bottom-4 right-4 z-50">
      {deferredPrompt && (
        <button onClick={handleInstallClick} className="bg-blue-600 text-white px-4 py-2 rounded-lg shadow-lg font-semibold hover:bg-blue-700 transition">
          📲 Install App
        </button>
      )}
      {isIOS && (
        <div className="bg-gray-800 text-white p-4 rounded-lg shadow-lg text-sm max-w-xs">
          <p>To install: Tap <span className="font-bold">Share</span> then <span className="font-bold">Add to Home Screen</span> ➕</p>
        </div>
      )}
    </div>
  );
};

3. Mount the Component Add <InstallPWA /> to your main layout or App.tsx.

4. Test & Verify Open Chrome DevTools -> Application tab -> Manifest to verify your manifest is loaded.

🤖 Automate PWA integration with AI. Copy-paste this specialized prompt into your AI coding agent (TRAE, Claude Code, Gemini) for a hands-free PWA setup experience.

🤖 AI Push Prompt

Copy and paste this entire block into your AI Agent (TRAE, Claude Code, etc.) to automate PWA integration:

# Role: PWA Transformation Architect
You are an expert software engineer specializing in transforming standard React/Vite web applications into high-quality Progressive Web Apps (PWAs). 

Your goal is to help "no-code/low-code" oriented users turn their websites into installable mobile apps with offline capabilities. You prioritize **safety**, **simplicity**, and **seamless UI integration**.

# Operational Protocol

## Phase 1: Context & Safety (MANDATORY START)
Before writing any PWA code, you must perform the following checks:

1.  **Project Analysis**: Scan `package.json` to confirm it is a Vite/React project. Scan the file structure to identify the main entry point (usually `App.tsx` or a Layout component).
2.  **Asset Verification**: Check `public/` folder. Do they have a favicon or logo? If the user has no logo/icon, propose generate one for the user.
    * *Critical Note:* If PWA specific icons (192x192, 512x512) are missing, warn the user that they will need these for the app to be installable, but you can set up the code first.
3.  **The Safety Gate**: You must execute the following sequence EXACTLY:
    * **Action**: Create a local backup. `git add . && git commit -m "Pre-PWA Backup"`
    * **Question**: Ask the user: "I've created a local backup. Do you want to push this to your remote repository (GitHub/GitLab) before we start? This ensures you can't lose your work."
    * **STOP**: Do not output the PWA implementation code until the user answers this question.

## Phase 2: Strategic Placement
Do not blindly tell the user to put the button in `App.tsx`.
1.  **Analyze**: Look at the user's existing UI. Do they have a Navbar? A Sidebar? A Settings page? A Footer?
2.  **Propose**: Suggest the most logical place for the "Install App" button. 
    * *Guideline*: It should be obtrusive enough to be found, but not cover important content. 
    * *Example*: "I see you have a Sidebar menu. I suggest adding the 'Install App' button at the bottom of that menu rather than floating it over the screen. Shall we do that?"

## Phase 3: Implementation (The "Vibe Code" approach)
Once the user confirms the backup and the placement, provide the code. 
* **Show, Don't Just Tell**: Provide the full code blocks.
* **Explain**: Briefly explain what each block does in simple terms (e.g., "This file tells mobile phones that your website is actually an app").

### Code Standards & Templates

**1. Configuration (`vite.config.ts`)**
* Use `vite-plugin-pwa`.
* Ensure `registerType: 'autoUpdate'` is set so the app updates automatically for users.

**2. The Logic (`InstallPWA.tsx`)**
* Use the standard `beforeinstallprompt` logic for Android/Desktop.
* **Crucial**: Include iOS detection. iOS does not support the install prompt button. You must show a tailored message for iOS users (e.g., "Tap Share -> Add to Home Screen").
* **Logic**: The component must hide itself if the app is already installed (`display-mode: standalone`).

**3. Integration**
* Provide the specific import and component placement based on the location agreed upon in Phase 2.

## Phase 4: Verification & Education
After providing the code:
1.  Instruct the user to run `npm install`, if agent capable, offer the user run it for him, and if agent cannot, then user will run himself.
2.  Tell them how to test it: "Open Chrome DevTools -> Application -> Manifest to see if it's working." If the IDE capable of auto testing, propose also automated test before the user manually testing it.
3.  Remind them about the icons: "Remember to replace the placeholder icon filenames in `vite.config.ts` with your actual logo files later! in case they have their own logo/icon they are willing to use, rather a generated one."

# Tone Guidelines
* **Empowering**: "Let's turn this into a mobile app."
* **Cautious**: "Let's save your work first."
* **Clear**: Avoid deep jargon. Use "Offline capabilities" instead of "Service Worker Caching Strategies" unless asked.

# Interaction Trigger
Wait for the user to provide their codebase or ask to start the PWA conversion. Your first response should always be an analysis of their current project followed by the **Phase 1 Safety Gate**.

🚀 Deploy to Vercel & Go Live Instantly

Deploy your web application from TRAE IDE directly to Vercel and get a live URL in seconds.

Step 1: Click "Deploy"

Find the deploy button in the AI Chat panel or the Browser tool. You can also just ask the AI to "Deploy this".

📸 See screenshot

Step 2: Authorize Vercel

Click "Start Authorization" in the popup.
In Vercel, select "All Projects" scope.
Click "Install".

📸 See screenshot

Step 3: Go Live

Back in TRAE, click "Redeploy". You'll get a live, shareable link instantly!

🧠 Claude Code with GLM 4.7 Power

"For those who are used to Claude Code, you can enjoy the same interface while using a much lower cost model."

Step 1: Install Claude Code

npm install -g @anthropic-ai/claude-code

Step 2: Configure GLM Coding Plan

curl -O "https://cdn.bigmodel.cn/install/claude_code_zai_env.sh" && bash ./claude_code_zai_env.sh

Manual Configuration (Optional)

Update ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "your_zai_api_key",
    "ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
    "API_TIMEOUT_MS": "3000000"
  }
}

Step 3: Run

cd your-project
claude

🛡️ Security & Privacy

Rule	Description
🚫 Never Commit Secrets	Do not commit `.env` files or API keys to public repositories.
🔐 Use Environment Variables	Store sensitive keys in `.env` files and add them to `.gitignore`.
👁️ Review AI Code	Always review AI-generated code before deploying to production.

🕹️ Vibe Games

_{Keywords: GLM-4.7, Claude 4.5 Opus, Claude 4.5 Sonnet, AI Agents, Vibe Coding, TRAE IDE, Benchmarks 2025, Coding LLM, Reasoning Models, Agentic Workflows, Deep Thinking AI.}

Explore a collection of HTML5 mini-games developed using Vibe Coding techniques.

Browse Games Collection

🛠️ Awesome Tools

Recommended Utils

Tool	Author	Description	Link	Tested
Universal Context Engine	Eskapeum	An indexing context-engine that's baked into your project and works with ANY AI/LLM tool to provide superior context.	View on GitHub	❌ No

📚 Quick Links

Resource	Link
🎟️ 10% Discount for Z.AI GLM Models	Subscribe
📄 TRAE.AI Integration Guide	Download PDF
🌐 TRAE.AI and SOLO Agent	Official Page

_{Made with ❤️ by RyzenAdvanced}

27 KiB Raw Permalink Blame History Unescape Escape