OneDayAgent: Towards a Long-Horizon Harness for Autonomous Agents
OneDayAgent is a harness for keeping autonomous agents on track across long, tool-heavy tasks.
The paper says it decomposes open-ended requests into bounded subtasks, preserves execution memory under context pressure, and checks and repairs the final output. It is evaluated on AgentIF-OneDay across 104 tasks. With GLM-5.2 as the backend, it reports an overall score of 0.821. The authors say the same harness runs across five LLM backends from three model families without tuning. HF Daily Papers' note
The paper says it decomposes open-ended requests into bounded subtasks, preserves execution memory under context pressure, and checks and repairs the final output. It is evaluated on AgentIF-OneDay across 104 tasks. With GLM-5.2 as the backend, it reports an overall score of 0.821. The authors say the same harness runs across five LLM backends from three model families without tuning. HF Daily Papers' note
score 5