ProcGrep
June 11, 2026
Agent trajectories as programs. Benchmark scores tell you what an agent got right; they do not tell you how it got there. We introduce methods for comparing agents procedurally, and find that ten agents are identifiable by their behavioral habits, which we define as fingerprints. ProcGrep is a library for auditing and evaluating agents for how they approach tasks at a procedural level given their traces.