让编程智能体只读关键代码片段，大幅减少冗余信息
Squeez: Task-Conditioned Tool-Output Pruning for Coding Agents
- 根据任务需求精准筛选工具输出中的必要内容
- 在保留92%关键信息的同时删除92%的输入文本
- 适合需要高效处理长代码的AI编程助手开发者
编程智能体频繁处理冗长的工具输出，但实际仅需其中一小部分来推进下一步。本文研究任务条件下的工具输出裁剪：给定一个聚焦问题和一条工具输出，返回智能体应继续检查的最小原文证据块。我们构建了一个包含11,477个样本的基准测试集，基于SWE-bench仓库交互与合成多生态工具输出生成，并人工标注了618个测试样本。通过LoRA微调Qwen 3.5 2B模型，并与更大规模的零样本模型及启发式裁剪基线对比。该模型达到0.86召回率和0.80 F1值，在移除92%输入令牌的同时，比零样本Qwen 3.5 35B A3B高出11个召回点，且显著优于所有启发式基线。
原文摘要 · Abstract (English)
Coding agents repeatedly consume long tool observations even though only a small fraction of each observation matters for the next step. We study task-conditioned tool-output pruning: given a focused query and one tool output, return the smallest verbatim evidence block the agent should inspect next. We introduce a benchmark of 11,477 examples built from SWE-bench repository interactions and synthetic multi-ecosystem tool outputs, with a manually curated 618-example test set. We fine-tune Qwen 3.5 2B with LoRA and compare it against larger zero-shot models and heuristic pruning baselines. Our model reaches 0.86 recall and 0.80 F1 while removing 92% of input tokens, outperforming zero-shot Qwen 3.5 35B A3B by 11 recall points and all heuristic baselines by a wide margin.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。