SPIN Unprocessed October 16, 2026 ai_technology research
Prune the Summary, Not the History: Source-Judged Compaction for Long-Horizon Agents
View original on arxiv.orgOverview
arXiv:2610.20927v1 Announce Type: new Abstract: Long-horizon LLM agents rely on context compaction: when the interaction history outgrows the context window, a compressor replaces it with a natural-language summary. We show that these summaries contain more than a record of what happened. Compaction prompts ask for decisions and next steps, so each summary mixes facts established by the environment with reasoning written by the compressor, and the agent cannot tell the two apart afterwards. In a
SpinGraph analysis pending — check back after processing.
Ask AI about this story
Opens with the SpinGraph .md URL and structured context — one click, prompt included.
More from arXiv Computation and Language
View all →- Does Morphological Correctness Matter for Tokenization? A Scrambled Control Study in Tagalog
- PEDBench: Can Speech Models Detect Phonetic Errors?
- CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning Distillation
- Wolof-Adapted Industry-Ready Open-Weight Language Models
- PRISM: Transfer and Limits of Activation Interventions in Tool-Calling Language Models
- APRiL: An Open-ended Benchmark for Chinese Social Awareness in Large Language Models
Markdown (.md) · JSON-LD schema (.json) · Machine-readable for AI & GEO