PrimusSaaS.Security.AI 2.4.10

PrimusSaaS.Security.AI

This package adds AI-powered remediation and deep security analysis capabilities to the PrimusSaaS framework. It leverages Microsoft.SemanticKernel to interact with Large Language Models (LLMs) like OpenAI, Azure OpenAI, or local models to suggest fixes for vulnerabilities.

Features

• AI Remediation: Automatically suggests code fixes for detected vulnerabilities.
• Model Agnostic: Works with any model supported by Semantic Kernel (OpenAI, Azure, HuggingFace, etc.).
• Reliable Parsing: Uses JSON-structured outputs to ensure consistent and parseable responses from the LLM.
• Safety Checks: Validates suggested fixes using the Roslyn compiler before offering them.

Installation

dotnet add package PrimusSaaS.Security.AI

Usage

You must configure a Kernel and inject it into the RemediationAgent.

using Microsoft.SemanticKernel;
using PrimusSaaS.Security.AI.Services;

var builder = WebApplication.CreateBuilder(args);

// 1. Configure Semantic Kernel
var kernelBuilder = Kernel.CreateBuilder();
kernelBuilder.AddOpenAIChatCompletion(
 modelId: "gpt-4",
 apiKey: "YOUR_OPENAI_API_KEY"
);
var kernel = kernelBuilder.Build();

// 2. Register the Kernel
builder.Services.AddSingleton(kernel);

// 3. Register the Remediation Agent
builder.Services.AddTransient<RemediationAgent>();

// Optional: Configure Options
builder.Services.AddTransient(sp => new RemediationOptions 
{
 Temperature = 0.2,
 RequireValidation = true
});

How it Works

1. Analyze: The agent receives a vulnerability report (Diagnostic ID, code snippet).
2. Prompt: It constructs a prompt asking the LLM to fix the specific issue and return a JSON object.
3. Process: The LLM generates a fix and an explanation.
4. Validate: The agent uses Roslyn to compile the suggested fix to ensure it is syntactically correct.
5. Result: Returns a RemediationSuggestion object.