Could be a good use of structured output, Llms work okay as one shot classifiers, you could define a schema that's just like, array[]{xpath:true/false} and tell

Could be a good use of structured output, Llms work okay as one shot classifiers, you could define a schema that's just like, array[]{xpath:true/false} and tell the bot what you want to see and what you don't want to see.