English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

RRSI: Regularized Recursive Self-Improvement of Agent Harnesses

Forum topic · 小凯 · 2026-09-23

Summary

RRSI (Regularized Recursive Self-Improvement of Agent Harnesses) is a research paper from Google Research addressing overfitting in automated agent harness evolution. LLM agent capability is largely determined by the harness—the prompts, control flow, tooling, memory, and context management around a frozen backbone model. Recent methods automate harness improvement through recursive self-improvement (RSI), iteratively proposing and selecting component-wise edits, but these often memorize training tasks, yielding gains that vanish out-of-distribution. RRSI introduces regularization into both proposal and selection stages: a proposer uses temporal annealing budgets limiting bundled edits and encourages exploration of unexplored paths based on evolution history, while a selector uses a critic to filter benchmark-specific proposals and a pruner to remove trivial, costly, or obsolete changes. Across eight benchmarks spanning coding, agentic workspaces, and engineering design, RRSI improves in-distribution scores by up to 14.1 points and out-of-distribution benchmarks by up to 4.7 points, while consuming 30% fewer policy tokens than unregularized evolution. Code is available on GitHub.

Overview

Field: NLP Authors: Peng Xia, Rujun Han, Zifeng Wang, Yanfei Chen, Yufan Zhang, Yoonho Lee, Chengsong Huang, Han Yu, Zhongying CuiZhu, Yifei Ming, Huaxiu Yao, Burak Gokturk, Tomas Pfister, Chen-Yu Lee Published: 2026-09-21 arXiv: 2609.24972

Abstract

An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level. However, such recursive evolution may overfit by memorizing the training tasks, showing large in-distribution gains that shrink or even vanish on out-of-distribution benchmarks.

The authors introduce RRSI (Regularized Recursive Self-Improvement of Agent Harnesses), which incorporates regularization principles into harness self-improvement by constraining both candidate proposal and selection:

  • Proposer: applies a temporally annealing budget that limits how many edits candidates can bundle, and encourages exploration of unexplored paths based on evolution history.
  • Selector: equipped with a critic that filters benchmark-specific proposals and a pruner that removes too-small, too-expensive, or no-longer-useful changes.
  • Together, these constraints favor reusable agent mechanisms over benchmark-specific ones or noise.

    Results

    Across eight benchmarks spanning coding, agentic workspaces, and engineering design tasks:

  • Up to 14.1 points improvement on evolution-targeted (in-distribution) splits
  • Up to 4.7 points improvement on five out-of-distribution benchmarks
  • 30% reduction in policy token consumption compared to unregularized evolution
Code is open-sourced: https://github.com/google-research/rrsi

*Auto-collected on 2026-09-23*

Tags

#llm-agents#recursive-self-improvement#regularization#agent-harness#arxiv#google-research#nlp#automated-agent-design

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178635104