CompactionRL Trains Long-Horizon Agents to Compress Their Own Context

arXiv·high signal

Long-horizon agent trajectories overflow the context window before tasks finish; CompactionRL uses reinforcement learning to teach the agent to summarize and compact prior interactions rather than naively truncating, preserving task-critical state. It targets the exact failure mode multi-step coding and research agents hit today. Directly actionable for anyone whose agent degrades once conversations get long.