ViberChecekr
This project is an AI-powered security scanner for GitHub repositories that uses Google's Gemini 2.5 Pro API and a three-agent architecture.
Core trick
Clones a GitHub repo into one payload, then runs a Gemini scanner, patcher, and validator loop up to three times with parser syntax checks.
Description
This project is an AI-powered security scanner for GitHub repositories that uses Google's Gemini 2.5 Pro API and a three-agent architecture.
A user submits a GitHub URL via a Flask interface. The system clones the repo and compiles its source code into a single payload.
Agent 1 (VulnerabilityScanner): Analyzes the payload using Gemini to detect four vulnerability types (hardcoded credentials, SQL injection, XSS, missing authentication), returning structured JSON findings.
Agent 2 (CodePatcher): If vulnerabilities are found, this agent receives the findings and the full codebase to generate minimal, secure code fixes that address the root cause.
Agent 3 (CodeValidator): Validates the proposed fixes in an evolutionary loop (up to 3 iterations). Validation is two-stage: Stage 1 checks syntax with parsers, and Stage 2 uses Gemini to confirm the fix works and doesn't introduce regressions, providing a final quality analysis.
README excerpt
# AI-Powered Security Vulnerability Scanner & Auto-Remediation System An intelligent security tool that uses Google's Gemini 2.5 Pro API to automatically detect and fix security vulnerabilities in GitHub repositories through a three-agent architecture. ## 🎯 Overview This project provides an end-to-end solution for identifying and remediating security flaws in codebases. Simply submit a GitHub repository URL, and the system will: 1. **Scan** for vulnerabilities using AI-powered analysis 2. **Generate** secure code fixes automatically 3. **Validate** fixes through an evolutionary loop with syntax checking and deep analysis ## 🏗️ Architecture ### Three-Agent System #### **Agent 1: VulnerabilityScanner** - Analyzes entire codebase using Gemini AI - Detects 4 critical vulnerability types: - Hardcoded credentials - SQL injection vulnerabilities - XSS (Cross-Site Scripting) issues - Missing authentication checks - Returns structured JSON findings with: - File paths and line numbers - Code snippets - Detailed security explanations #### **Agent 2: CodePatcher** - Generates secure, production-ready code fixes - Uses minimal changes to preserve functionality - Supports both individual and batch fixes - Addresses root security issues comprehensively #### **Agent 3: CodeValidator** - Validates fixes through evolutionary loop (up to 3 iterations) - **Stage 1**: Language-specific parser validation for syntax correctness - **Stage 2**: Deep AI analysis for security effe