Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectories

ArXi:2604.17596v1 Announce Type: cross We release Terminal Wrench, a subset of 331 terminal-agent benchmark environments, copied from the popular open benchmarks that are nstrably reward-hackable. The data set includes 3,632 hack trajectories and 2,352 legitimate baseline trajectories across three frontier models (Claude Opus 4.6, Gemini 3.1 Pro, GPT-5.4). Each entry preserves the original task definition alongside full attack trajectories that show how the verifier was bypassed. It also includes cases where the task was not solved as intended.