SKILL.state: Scalable Long-Horizon Agent Skills
Sanket Badhe, Priyanka Tiwari and Jonghyun Chung (EMNLP) replace the append-only conversation history in agent runtimes with an explicit mutable execution state, so prompt size stops tracking execution length.
Ask this paper
What the model sees each step: Only the immutable skill specification, the current structured execution state, and the latest observation. Nothing else carries forward.
Reasoning is discarded on commit: Intermediate reasoning is thrown away immediately after it produces a validated state update, which is the mechanism that prevents prompt growth and context poisoning over long horizons.
Accuracy up, tokens down: Across datasets, models and execution environments, task accuracy improves while cumulative token consumption drops substantially, so this is not the usual accuracy-for-cost trade.
Architecture-agnostic: The claim is that explicit execution state is a runtime abstraction, not a model-specific trick, which makes it directly portable to existing skill runtimes.
Why it matters: Long-running agents currently degrade for two coupled reasons, latency and context poisoning, and both are downstream of append-only history. Making state explicit and mutable addresses the shared cause.
Abstract
Large Language Models (LLMs) increasingly act as autonomous agents executing complex, long-running procedural skills. Existing agent runtimes maintain execution by continually appending observations, actions, and intermediate reasoning traces to an ever-growing conversation history, causing latency degradation and context-poisoning failures over long horizons. We present SKILL.state, a runtime architecture that replaces append-only conversational history with an explicit, mutable execution state. At each execution step, the model receives only the immutable skill specification, the current structured execution state, and the latest observation. Intermediate reasoning is discarded immediately after producing a validated state update, preventing prompt growth with execution history. Across diverse datasets, models, and execution environments, SKILL.state improves task accuracy while substantially reducing cumulative token consumption. Our results demonstrate that explicit execution state is an effective and architecture-agnostic abstraction for scalable long-horizon agent skills.