Context as a Tool: Context Management for Long-Horizon SWE-Agents

TL;DR

This paper introduces CAT, a new context management system for long-horizon software engineering agents, which improves reasoning stability and scalability by proactive context compression and structured memory, outperforming existing methods.

Contribution

The paper presents a novel context management paradigm called CAT, including a formal structure and a training framework, significantly enhancing long-term reasoning in SWE-agents.

Findings

SWE-Compressor achieves a 57.6% success rate on SWE-Bench-Verified.

Outperforms ReAct-based agents and static compression baselines.

Maintains stable, scalable reasoning under bounded context budgets.

Abstract

Agents based on large language models have recently shown strong potential on real-world software engineering (SWE) tasks that require long-horizon interaction with repository-scale codebases. However, most existing agents rely on append-only context maintenance or passively triggered compression heuristics, which often lead to context explosion, semantic drift, and degraded reasoning in long-running interactions. We propose CAT, a new context management paradigm that elevates context maintenance to a callable tool integrated into the decision-making process of agents. CAT formalizes a structured context workspace consisting of stable task semantics, condensed long-term memory, and high-fidelity short-term interactions, and enables agents to proactively compress historical trajectories into actionable summaries at appropriate milestones. To support context management for SWE-agents, we propose a trajectory-level supervision framework, CAT-GENERATOR, based on an offline data construction pipeline that injects context-management actions into complete interaction trajectories. Using this framework, we train a context-aware model, SWE-Compressor. Experiments on SWE-Bench-Verified demonstrate that SWE-Compressor reaches a 57.6% solved rate and significantly outperforms ReAct-based agents and static compression baselines, while maintaining stable and scalable long-horizon reasoning under a bounded context budget.