and then there's the whole "well MCP!" and in theory that's great....
....if you discount "your llm token usage blows through your budget immediately when someone discovers how to prompt inject the llm instance doing your frontend and suborns it for their openclaw botnet" as a fucking threat.
or, as annotated above, having the thing give true/false positivity determinations that get used by other systems that could then open up other systemic avenues of attack because holy shit security is a -systems- problem and you have to maintain an understanding of how all the parts work together to properly evaluate the impact of a given fucking vulnerability fucking surface.
