A bot that scrapes network requests for a website and detects security vulnerabilities using Code Llama
This was coded on my Youtube live stream
The script runs a puppeteer instance that intercepts all http requests from page load. After the page load the script fills in a random form and submits it (simulating user data over the wire).
After this we get all the intercepted requests and format them for LLM prompting (chunk the requests by 10 for context window optimization). In the end we send the prompt to Code Llama through an API I wrote in Colab (see below).
(if you want to change this and use llama.cpp please make a github issue and I'll address or message me on how to refactor for yourself) The bot runs off of an API coded in colab that's running flask/ngrok. The API is in front of Code LLama. You can run the colab here. Remember you have to copy the ngrok url to the 10x-hacker-bot file
Update this url with the new ngrok url from flask
const apiUrl = "http://8c54-34-87-182-84.ngrok.io"; // change this line
First npm install to get the node_modules directory
Running 10x-hacker-bot.js
$ node 10x-hacker-bot.js <website>
You can choose to run the API from the colab I have available which is a simple Flask app or simply download the code from the colab and run it locally if you'd prefer (if you have a quality GPU)