hack-verifiable-environments/hv-terminal-bench-2-1

Hack-verifiable version of terminal-bench-2-1. Each task includes hidden solution and test files; inotifywait detects whether agents access them instead of solving legitimately.

← Back to Registry

Run this task

CLI:

inspect eval inspect_harbor/hack_verifiable_environments_hv_terminal_bench_2_1 --model openai/gpt-5

Python:

from inspect_ai import eval
from inspect_harbor import hack_verifiable_environments_hv_terminal_bench_2_1

eval(hack_verifiable_environments_hv_terminal_bench_2_1(), model="openai/gpt-5")

Dataset information

Harbor registry	hack-verifiable-environments/hv-terminal-bench-2-1
Inspect task	hack_verifiable_environments_hv_terminal_bench_2_1
Latest digest	sha256:f77c76cd022b0aec434bc755ac1cfd8da650237c72047f189747fa10e33f6907
Samples	89
Paper	arxiv
Source	https://github.com/MajoRoth/hack-verifiable-environments

See Task Parameters for the parameter set shared across all Harbor tasks.