How to build an adversarial eval for your agent cache (and why 18 pairs matter more than a benchmark)

· Dev.to