X-Tree Recovers Hierarchical Action Structures for Agent Generalization
September 25, 2026
X-Tree tokenizes reusable agent experiences by scoring action spans for reusability and merging them into a hierarchical tree. This allows agents to learn sub-procedures directly into their weights without requiring LLM-driven skill retrieval in context.
HOW THIS AFFECTS YOU
●
builderYou can improve agent generalization by training on hierarchical routines rather than flat action streams.
●
researcherThis presents a way to move structured planning from the prompt context into the model weights.