Configurable agent benchmark with AI judge reports
Agent benchmark matrix — task × plugin combinations