Brilliant use case - to what degree is the agent being guided? Is it mostly a pairing exercise or agent exploration followed by human validation?

I pointed it at the library's git and let it run from there! I validated all the output (vuln description, risk assessment, exploit poc, patch) before reporting to maintainers

It is surreal to see a zero day like this being caught entirely by AI. What was the toughest part about building this?

Also, a clever title.