pkg/aflow/flow/assessment: question LLM confidence better #6782
Description
Out of 170+ runs LLM never said it's not confident (perhaps because we asked it to be a "kernel expert",
and experts can't be non-confident :)). Either way the Confident flag is not working as intended.
Perhaps we a second agent "skeptic" agent that won't consider the initial assessment as "own",
and will be able criticize it. The second agent could review the explanation/benign-ness for correctness
and confidence and produce the Confident flag.