Hugging Face has published a research paper titled "Scaling Agentic Data for Long-Horizon Terminal Intelligence," which addresses a central challenge in building terminal agents capable of sustained autonomous work. According to the paper, short trajectories can demonstrate commands and tool calls, but they fail to capture how agents accumulate evidence through exploration, adjust strategies based on feedback, and convert intermediate results into final deliverables.
The authors argue that high-quality long-horizon trajectories are needed to capture these agentic behavioral patterns and their connections within a single problem-solving process. However, they note that environment synthesis may complete or simplify the key work underlying these patterns in advance, removing the exploration, implementation, and verification the agent was meant to perform.
To address this, the paper introduces TerminalHorizon, described as a fully automated data synthesis engine for long-horizon terminal agents. Starting from real-world professional work, it uses independent execution to identify key work omitted during construction and uses the resulting evidence to guide environment reconstruction. The engine also progressively expands tasks under the same goal, introducing new requirements that build on prior results and call for further exploration and improvement.
Using this engine, the authors construct 1.5K task environments and perform supervised fine-tuning of Qwen3.5-27B on the resulting dataset to obtain TerminalHorizon-27B. They report substantial performance gains on all 12 benchmarks spanning coding-agent tasks, agentic tasks in real-world production settings, and general reasoning.
The paper also describes multiple RSI case studies, each with a 12-hour research budget, which further demonstrate the model's ability to continue experimenting beyond an initial solution, improve its methods based on feedback, and retain effective improvements in its final deliverables.