Jingyu Xiao, Qianxi Liu, Yintong Huo, Yuan Jiang, Michael R. Lyu · 2026 · arXiv
DOI: 10.48550/arXiv.2610.28612 · open-access PDF
Software-Engineering-Agenten lösen Repository-Level-Aufgaben durch iterative Werkzeugnutzung, aber ihre kontinuierlich wachsenden Interaktionsspuren erleiden erheblichen Token-Verbrauch und Schlussfolgerungen. Bestehende Komprimierungsansätze erfordern oft Modelltraining, Hilfs-LLM-Aufrufe oder Annahmen über spezifische Inhalte wie Quellcode. Wir präsentieren TraceLite, ein einfaches, aber wirksames, ausbildungsfreies Rahmenwerk für die Komprimierung heterogener SWE-Agent-Beobachtungen.
Machine translation (Meta M2M100); the English original follows.
Software engineering agents solve repository-level tasks through iterative tool use, but their continuously growing interaction traces incur substantial token consumption and inference overhead. Existing compression approaches often require model training, auxiliary LLM calls, or assumptions about specific content such as source code. We present TraceLite, a simple yet effective, training-free framework for compressing heterogeneous SWE-agent observations. Our preliminary study shows that optical rendering preserves substantial textual and semantic information under moderate compression, while
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.28612, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange