Self-Healing Harness for Runtime Oversight of Agent Self-Modification

2026-09-21

AI Summary

研究提出一种自愈框架，通过检测、通知、修复、验证循环来控制LLM代理的自我修改，以防止局部有益修改引入副作用。

Abstract

LLM agents can change their own future behavior, raising a basic control question of which self-generated changes should be allowed to persist. We formulate this as admission control for self-modification. The agent may propose changes to its operating instructions, while an external runtime gate controls persistence. We implement this principle as a model-agnostic self-healing harness that runs a Detect, Notice, Heal, Validate loop around an otherwise unmodified agent. The agent authors candidate behavioral rules in an external workspace, where they receive provisional execution authority during evaluation and acquire persistent cross-episode authority only after measured improvement on the triggering failure without regression beyond a fixed margin on protected cases. Replay provides matched evidence when available, forward trials provide a weaker fallback, and a corpus-level guard re-tests the accumulated active rule set. Across 16 matched Baseline and Harness runs spanning AppWorld, Terminal-Bench, and $τ^2$-Bench, the gate rejected 383 replay-decided proposals. Of these, 211 (55%) improved their triggering failure while degrading a case that previously worked. This shows that locally beneficial self-modifications can introduce collateral regressions often enough to materially affect gate decisions, providing direct empirical motivation for external admission control. Task-completion score is higher under the Harness in all 16 pairs, with two paired bootstrap intervals excluding zero, while repeated-trial reliability is higher in 12 pairs, tied in 4, and lower in none. Because adaptation modifies the policy-inducing context while leaving model weights fixed, admitted changes remain inspectable, reversible, and compatible with closed-weight models.

Problem

• Self-Modification
• Admission Control
• Runtime Oversight
• Agent Behavior
• External Gate

Innovation

• self-healing harness
• admission control for self-modification
• runtime oversight
• behavioral rules evaluation
• external workspace