Builder

harbor-framework

1 Tool

Harbor is an open-source framework for evaluating and improving AI agents, built around a Dockerfile-plus-instructions-plus-test task format. Its flagship project, Terminal-Bench, benchmarks AI agents on complex, realistic command-line tasks, and the team has followed it with Frontier-Bench, a related evaluation effort. The project's public leadership is credited to Ryan Marten. Harbor is released under Apache 2.0 and is written primarily in Python.