SWE-Pruner: Self-Adaptive Context Pruning for Coding Agents

TL;DR

SWE-Pruner is a self-adaptive, task-aware context pruning framework for coding agents that dynamically selects relevant code segments, reducing context size significantly while maintaining or improving performance.

Contribution

It introduces a neural skimmer guided by explicit task goals to perform adaptive pruning, addressing limitations of fixed-metric compression methods in code understanding tasks.

Findings

Achieves 23-54% token reduction on multiple benchmarks.

Up to 14.84x compression on single-turn tasks with minimal performance loss.

Improves success rates in code-related tasks.

Abstract

LLM agents have demonstrated remarkable capabilities in software development, but their performance is hampered by long interaction contexts, which incur high API costs and latency. While various context compression approaches such as LongLLMLingua have emerged to tackle this challenge, they typically rely on fixed metrics such as PPL, ignoring the task-specific nature of code understanding. As a result, they frequently disrupt syntactic and logical structure and fail to retain critical implementation details. In this paper, we propose SWE-Pruner, a self-adaptive context pruning framework tailored for coding agents. Drawing inspiration from how human programmers "selectively skim" source code during development and debugging, SWE-Pruner performs task-aware adaptive pruning for long contexts. Given the current task, the agent formulates an explicit goal (e.g., "focus on error handling") as a hint to guide the pruning targets. A lightweight neural skimmer (0.6B parameters) is trained to dynamically select relevant lines from the surrounding context given the goal. Evaluations across four benchmarks and multiple models validate SWE-Pruner's effectiveness in various scenarios, achieving 23-54% token reduction on agent tasks like SWE-Bench Verified while even improving success rates, and up to 14.84x compression on single-turn tasks like LongCodeQA with minimal performance impact.