PromptAgent

Prompt 優化測試系統 — 使用 Microsoft Agent Framework 進行智慧評估與多輪優化

PromptAgent 是一個 Prompt 品質測試與優化工具。透過多輪平行執行與 AI 評估，快速找到最佳 Prompt 配置。

Features

• Multi-Agent Parallel Execution — 使用 Task.WhenAll 同時執行多個 Agent，測試 Prompt 穩定性
• GAI Feasibility Analysis — 智慧分析需求適合 AI、傳統程式、還是人工處理
• Prompt Version Control — 像 Git 一樣管理 Prompt 版本，支援回滾與 Diff 比較
• Smart Correction Mode — 全自動多輪優化，達標自動停止
• Guided Mode — 每輪暫停讓用戶選擇偏好的回答風格
• AI Example Generation — LLM 動態生成測試範例

Test Modes

Mode	Description	Use Case
Normal	Single execution + AI evaluation	Quick validation
Smart Correction	Fully automatic multi-round	Trust AI judgment
Guided	Pause each round for user input	Precise control

Quick Start

1. Clone

git clone https://github.com/breezy89757/PromptAgent.git
cd PromptAgent

2. Configure

cp appsettings.template.json appsettings.json

Edit appsettings.json:

{
  "AzureOpenAI": {
    "Provider": "Azure",
    "Endpoint": "https://your-resource.openai.azure.com/",
    "ApiKey": "your-api-key",
    "DeploymentName": "gpt-4o-mini",
    "EvaluatorDeploymentName": "gpt-4o"
  }
}

3. Run

dotnet run

Open http://localhost:5036

Project Structure

PromptAgent/
├── Models/           # Data models
├── Services/         # Business logic (Agent, Evaluation, Versioning)
├── Components/       # Blazor pages
└── appsettings.json  # Configuration

Data Storage

All data is stored in browser LocalStorage:

Feature	Key	Limit
Test History	prompt_test_history	30 records
GAI Evaluation	gai_evaluation_history	20 records
Version Projects	prompt_projects	Unlimited

Note

Clearing browser data will delete these records.

License

MIT