Terminal Agents Survey Argues Realized Agent Behavior Is Set by Harness and Runtime, Not the Model Alone

arXiv 2608.20485·medium signalXBlueskyLinkedInCopy link

Terminal Agents (arXiv 2608.20485, Aug 20) is a survey defining terminal agents as systems whose dominant progress-bearing action-observation loop runs through command execution, textual feedback and stateful environment interaction, a category currently scattered across software engineering, tool use and computer-use research. It organizes the space through a seven-dimensional terminal competence profile and concludes that behavior is jointly shaped by model, interface, harness, runtime and environment, with prevailing evaluations emphasizing final outcomes and exposing process quality, recovery and governance unevenly. The concrete recommendation is explicit reporting of system and runtime conditions with replayable traces, since matched system comparisons showed benchmark-dependent performance and limits on component attribution.