Search papers, labs, and topics across Lattice.
This paper introduces SkillGLoW, a novel approach for consolidating skills in self-improving agents that addresses the limitations of existing methods when tackling long-horizon tasks. By aggregating local skills into procedural families and compressing them into global priors, SkillGLoW enables agents to reuse effective solving procedures while minimizing the storage of instance-specific details. The method demonstrates significant performance improvements across multiple benchmarks, achieving an average gain of 17.2 points over a no-skill baseline and outperforming traditional single-document optimizers in most scenarios.
SkillGLoW reveals that procedural families, rather than task-specific memories, are the key to enhancing performance in long-horizon task streams.
LLM agents increasingly self-improve by writing and reusing textual skills, kept either as one global document or as a flat pool of per-task entries, though most of the evidence comes from domains with structurally similar tasks. On long-horizon workloads where each task demands a different solution, the two forms fail in opposite ways: the document collapses into generic discipline, while the pool inflates and its entries stay bound to the instance that wrote them. We argue the missing unit of reuse is the solving procedure shared by a cluster of related tasks, and build SkillGLoW (Global-Local Weave) around it: the local skills a task writes from its own execution are aggregated into procedural families and compressed into de-instantiated global priors, while the instance detail they hold is regenerated per task rather than stored; a commit gate admits a prior only when real execution shows it does not degrade the deployed library. Across four benchmarks (mathematical reasoning, terminal automation, software repair, and embodied control) and three models, the priors gain 17.2 points (hard) over the no-skill baseline on average, with positive gains in all 12 continual-improvement runs, and 18.0 with local regeneration, while the library holds one prior per procedural family, 3.6x more compact than the per-task pool. Under the same protocol GLoW leads a published single-document optimizer on 15 of 21 cells. Unmodified, the library lifts success on unseen ALFWorld tasks from 73.9% to 83.9%, evidence that what transfers is procedure rather than task memory.
