SABER: Small Actions, Big Errors -- Safeguarding Mutating Steps in LLM Agents

TL;DR

This paper analyzes the impact of mutating actions on LLM agent failures, introduces a safeguard method called extsc{Saber} to improve robustness, and releases a revised benchmark to better evaluate long-horizon tasks.

Contribution

It provides an action-level analysis of failure causes in LLM agents, introduces a novel safeguard method extsc{Saber}, and releases an improved benchmark dataset.

Findings

Mutating actions significantly impact success rates, deviations reduce success odds by up to 96%.

extsc{Saber} improves performance by up to 28% on benchmark tasks.

Ceiling effects in existing benchmarks are addressed with the new extsc{tau}-Bench Verified dataset.

Abstract

Despite rapid progress in LLM agents, performance on long-horizon, tool-using tasks remains fragile. To better understand this fragility, we ask a simple question: \emph{do all actions contribute equally to failure?} Analyzing execution traces on $\tau$-Bench (Airline/Retail) and SWE-Bench Verified, we decompose trajectories into \emph{mutating} (environment-changing) vs.\ non-mutating steps and formalize \emph{decisive deviations}, earliest action, level divergences that flip success to failure. A logistic regression reveals that each additional deviation in a mutating action reduces the odds of success by upto $92\%$ on Airline and upto $96\%$ on Retail for SoTA models. In contrast, deviations in non-mutating actions have little to no effect. Errors also grow with context length as agents drift from role and act on stale constraints. Motivated by these observations, we introduce \cm{}, a model-agnostic, gradient-free, test-time safeguard that (i) adds mutation-gated verification, (ii) injects \emph{Targeted Reflection} before mutating steps, and (iii) performs block-based context cleaning. \cm{} delivers consistent gains, e.g., Qwen3-Thinking: +28\% \emph{relative} on Airline, +11\% on Retail, and +7\% on SWE-Bench Verified; Claude: +9\%/+7\%. We further identify ceiling effects in $\tau$-Bench, where annotation errors and underspecified tasks artificially cap model performance. To address this, we release $\tau$-Bench Verified, which restores benchmark headroom through targeted revisions. Our results argue for action-level analysis, targeted safeguards, and reliable evaluations as prerequisites for robust multi-turn agents.