Agent Skills are structured packages of procedural knowledge that augment large language model (LLM) agents at inference time. Despite rapid adoption, there is no standard way to measure whether they actually help. We present SkillsBench, a benchmark whose current inventory contains 87 tasks across 8 domains paired with curated Skills and deterministic verifiers. Our latest aggregate evaluation runs the 87-task benchmark under matched no-Skills and curated-Skills conditions for 18 model-harness configurations. Curated Skills raise the average pass rate from 33.9
Large Language Model (LLM) agents have shown stunning results in complex tasks, yet they often operate in isolation, failing to learn from past experiences. Existing memory-based methods primarily store raw trajectories, which are often redundant and noise-heavy. This prevents agents from extracting high-level, reusable behavioral patterns that are essential for generalization. In this paper, we propose SkillRL, a framework that bridges the gap between raw experience and policy improvement through automatic skill discovery and recursive evolution. Our approach introduces an experience-based distillation mechanism to build a hierarchical skill library SkillBank, an adaptive retrieval strategy for general and task-specific heuristics, and a recursive evolution mechanism that allows the skill library to co-evolve with the agent's policy during reinforcement learning. These innovations significantly reduce the token footprint while enhancing reasoning utility. Experimental results on ALFWorld, WebShop and seven search-augmented tasks demonstrate that SkillRL achieves state-of-the-art performance, outperforming strong baselines over 15.3
We review the key observations and theories relevant to the internal structure and dynamics of the Galilean satellites. Key observations include: the bulk densities and degree-two gravity coefficients of the moons; the presence of conductive subsurface layers, as inferred from magnetic induction; and the surface compositions. All the moons, with the possible exception of Callisto, appear to be differentiated (denser components have separated from lighter components). Ganymede and Io have iron cores; Europa may have one. The outer three moons all likely possess subsurface oceans; for Europa the ocean overlies rock, while for the other two it overlies higher-pressure ice phases. Io is partially molten but does not possess a shallow magma ocean. Tidal heating is the dominant energy source at Europa and Io, and may have affected Ganymede’s long-term evolution. The dynamics of the subsurface oceans are of considerable theoretical interest but are only weakly tied to current or likely future observations. We identify seven outstanding questions regarding internal structures, some of which will be answered by the forthcoming JUICE, Europa Clipper and Tianwen-4 missions.
We review the key observations and theories relevant to the origin and evolution of the Galilean satellites. Key observations include: the potentially undifferentiated nature of Callisto; the increasing ice fraction with semi-major axis; the present-day existence of the Laplace resonance; the potential resurfacing of Ganymede mid-way through its evolution; and the metal-enriched nature of Jupiter’s envelope. The most widely accepted theory for the formation of the satellites is the so-called “starved disk” model, although newer alternatives including decretion disks and pebble accretion have also been proposed. Models that allow slow satellite formation in a cold disk are preferred, based on the density progression and Callisto’s apparent differentiation state. Major model uncertainties include the angular momentum distribution of the material infalling to the circumplanetary disk, the source of the solids, and the thermal and viscosity structure of the disk. We identify six outstanding questions, some of which will be answered by JUICE, Europa Clipper and Tianwen-4. A major difficulty in answering some questions is overprinting of primordial characteristics by later events.
To support long-term interaction in complex environments, LLM agents require memory systems that manage historical experiences. Existing approaches either retain full interaction histories via passive context extension, leading to substantial redundancy, or rely on iterative reasoning to filter noise, incurring high token costs. To address this challenge, we introduce SimpleMem, an efficient memory framework based on semantic lossless compression. We propose a three-stage pipeline designed to maximize information density and token utilization: (1) Semantic Structured Compression, which distills unstructured interactions into compact, multi-view indexed memory units; (2) Online Semantic Synthesis, an intra-session process that instantly integrates related context into unified abstract representations to eliminate redundancy; and (3) Intent-Aware Retrieval Planning, which infers search intent to dynamically determine retrieval scope and construct precise context efficiently. Experiments on benchmark datasets show that our method consistently outperforms baseline approaches in accuracy, retrieval efficiency, and inference cost, achieving an average F1 improvement of 26.4% in LoCoMo while reducing inference-time token consumption by up to 30×, demonstrating a superior balance between performance and efficiency.
