OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents

OneDayAgent is a long-horizon harness for autonomous agents handling open-ended, cross-environment, multimodal everyday tasks. It decomposes requests into bounded subtasks, maintains execution memory under context pressure, and verifies and repairs final deliverables. On 104 tasks from AgentIF-OneDay, the harness reportedly achieved an overall score of 0.821 with the GLM-5.2 backend, described as a new state of the art. It also ran across five backend LLMs from three model families without backend-specific tuning, although the abstract notes that different models produced distinct execution styles.