---
title: "从嘈杂轨迹到根因：面向Agent优化的结构轨迹分析与因果提取"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.07702v1"
permalink: "https://feed.astralor.com/items/itm_7f2f393f535edcbb7cee756d"
published: "2026-07-08T17:57:27.000Z"
score: 77
channel: "tech"
tags: ["Agent优化","STRACE","因果提取","形式验证","故障模式挖掘"]
---

# 从嘈杂轨迹到根因：面向Agent优化的结构轨迹分析与因果提取

## AI 点评

> 这篇论文直击Agent优化中“噪声轨迹导致优化失效”的痛点，STRACE的因果定位方法非常巧妙。对于正在开发自主Agent的团队或研究者，这是近期最值得关注的技术之一——尤其是那些面临复杂多步任务、调试困难的朋友，建议仔细阅读论文并尝试复现代码。

## 速览

长周期Agent优化常依赖基于LLM的反思机制，但真实执行轨迹存在冗余和异构，导致优化低效且易过拟合。STRACE框架通过批次级故障模式挖掘和轨迹内因果定位，构建高信噪比优化上下文。在VeruSAGE-Bench形式验证任务上，将人类专家设计的Agent成功率从42.5%提升至58.5%（提升1.4倍），代码已开源。

## 关键点

- STRACE通过批次级冗余轨迹过滤和代表性故障保留，解决大规模轨迹集合的异构与冗余问题。
- 在单条轨迹内，基于文本依赖图进行因果定位，去除无关步骤并识别真正根因模块。
- 在VeruSAGE-Bench基准上，STRACE显著超越标准上下文过滤基线，成功率提升1.4倍。
- 框架已开源，适用于长周期Agent的自动化调试与优化。

## 全文

The optimization of long-horizon agents increasingly relies on reflection-based mechanisms, where a large language model (LLM) acts as an optimizer to diagnose agent failures and improve agent policies. However, real execution traces are difficult to use directly for optimization: large trace collections are often redundant and heterogeneous, making optimization inefficient and prone to overfitting to low-value failures; meanwhile, each individual trajectory also contains many irrelevant steps, while naive context reduction methods such as truncation or sliding windows can discard causally important evidence and produce misleading optimization signals. To resolve this dilemma, we introduce STRACE (Structural TRajectory Analysis and Causal Extraction), a framework that constructs high signal-noise optimization contexts for more precise and effective optimization. At the batch level, STRACE mines failure patterns to filter redundant traces and retain representative failures; within each selected trace, it performs causal localization over a textual dependency graph to remove non-causal steps and identify the true root-cause module for optimization. Empirical results demonstrate that STRACE significantly outperforms standard context-filtering baselines. Notably, on a challenging formal verification task (VeruSAGE-Bench), it successfully optimizes human-expert designed agents, delivering $1.4\times$ success-rate improvement (42.5% to 58.5%). The code is available at https://github.com/moomight/STRACE .

---

引用前请核对原文;本文由星际流动整理:https://feed.astralor.com/items/itm_7f2f393f535edcbb7cee756d