Agent testing needs an LLM evaluator, not static scripts
With 276 commits, this framework uses an LLM evaluator to test agent reasoning via multiturn dialogue instead of static string matching.
Diego Alvarez
12 min read