Context Engineering | Compaction & Agent Memory for Automated Malware Analysis

SentinelLABS evaluated OpenAI’s native Responses API compaction for long-running automated malware analysis and found compaction reduced input tokens by ~86% with no measurable change to the aggregate evaluation score, while occasionally degrading higher-level domain object modeling; the report details methodology, results, server-side vs standalone compaction patterns, and practical guidance to separate working memory from durable artifacts.