Open sourcePython · ≈6.6k stars · Apache-2.0
Harbor
Framework for evaluating and improving agents
Harbor is a framework for evaluating and improving agents, from the harbor-framework organization that also maintains Terminal-Bench.
In research (9)
Oct 27, 2026 · Harbor ships Terminus-2, the top-scoring harness in Table 2.
- AutoRSE: Evaluating LLM Agents as Recommendation Engineers for Self-Evolving Recommender SystemsVerdict: Useful
Oct 27, 2026 · Harbor provides Terminus-2, the third harness evaluated.
- kw-bench: Breaking Frontier Harnesses on Real-World TasksVerdict: Useful
Oct 26, 2026 · The benchmark is packaged as Harbor tasks (Harbor 0.24.0).
- Work Agents: Systems, Capabilities, and OptimizationVerdict: Incremental
Oct 26, 2026 · Evaluation harness for isolated-container tasks.
- RepoTrace: Improving Requirement Tracing in Multi-Round Agent InteractionsVerdict: Incremental
Oct 26, 2026 · The Harbor agent environment and its Terminus-2 harness are used to run the evaluation.
- SWE-DIVE: Multi-Turn Coding Agent Evaluation with Dynamic User SimulationVerdict: Incremental
Oct 21, 2026 · SWE-DIVE is implemented as a Harbor environment.
Showing the 6 newest of 9 papers.
Neighbours on the workbench
Go · ≈49k stars · Apache-2.0
alibaba/open-code-review
Alibaba's hybrid deterministic-plus-agent code review CLI.
Maintained by Alibaba
Python · ≈39k stars · MIT
stanfordnlp/dspy
Programming, not prompting: optimize the whole pipeline.
Maintained by Stanford NLP
TypeScript · ≈37k stars · Other
langfuse/langfuse
Open-source LLM engineering platform: traces, evals, prompts.
Maintained by Langfuse
Connections
Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.
Sources
Verified Oct 29, 2026