ReAct+DT: A Self-Evolving Autonomous Agent Framework via Skill Distillation and Hierarchical Tree Retrieval

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Author: Wang, Zhongren
Format: Recurso digital
Published: Zenodo 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901984997539840
author Wang, Zhongren
author_facet Wang, Zhongren
contents <p>Current Large Language Model (LLM) agents primarily rely on the ReAct (Reason + Act) paradigm for task execution. However, these agents often suffer from "episodic amnesia," where successful reasoning chains are discarded after task completion, leading to redundant computation and high token latency in recurring scenarios. We propose <strong>ReAct+DT</strong>, an advanced framework that introduces <strong>Deposition (D)</strong> and <strong>Tree-structured Retrieval (T)</strong>. By distilling successful reasoning paths into permanent Python scripts and Markdown manuals, and organizing them into a self-growing hierarchical tree, the agent transitions from unstable zero-shot reasoning to stable, programmatic skill execution. Experimental results demonstrate that ReAct+DT significantly reduces inference costs and improves long-term task success rates.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18869913
institution Zenodo
language
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle ReAct+DT: A Self-Evolving Autonomous Agent Framework via Skill Distillation and Hierarchical Tree Retrieval
Wang, Zhongren
Autonomous Agents
ReAct Paradigm
Self-Evolving Systems
Skill Distillation
Hierarchical Retrieval
Long-term Memory
<p>Current Large Language Model (LLM) agents primarily rely on the ReAct (Reason + Act) paradigm for task execution. However, these agents often suffer from "episodic amnesia," where successful reasoning chains are discarded after task completion, leading to redundant computation and high token latency in recurring scenarios. We propose <strong>ReAct+DT</strong>, an advanced framework that introduces <strong>Deposition (D)</strong> and <strong>Tree-structured Retrieval (T)</strong>. By distilling successful reasoning paths into permanent Python scripts and Markdown manuals, and organizing them into a self-growing hierarchical tree, the agent transitions from unstable zero-shot reasoning to stable, programmatic skill execution. Experimental results demonstrate that ReAct+DT significantly reduces inference costs and improves long-term task success rates.</p>
title ReAct+DT: A Self-Evolving Autonomous Agent Framework via Skill Distillation and Hierarchical Tree Retrieval
topic Autonomous Agents
ReAct Paradigm
Self-Evolving Systems
Skill Distillation
Hierarchical Retrieval
Long-term Memory
url https://doi.org/10.5281/zenodo.18869913