Robustness tests for tool-using agents now check whether the task survived the edit: rescoring one trajectory set drops Mistral's attacked tasks from 64 to 26
Robustness tests for tool-using agents can now check, before the agent runs, whether an edited test case still tests the same task; rescoring the same 14,336 stored trajectories under executable task contracts drops Mistral's attacked-task count from 64 to 26 under a MiniLM cosine threshold.