Towards Secure Agent Skills: Architecture, Threat Taxonomy, and Security Analysis

Agent Skills is an emerging open standard that defines a modular, filesystem-based packaging format enabling LLM-based agents to acquire domain-specific expertise on demand. Despite rapid adoption across multiple agentic platforms and the emergence of large community marketplaces, the security properties of Agent Skills have not been systematically studied. This paper presents the first comprehensive security analysis of the Agent Skills framework. We define the full lifecycle of an Agent Skill across four phases -- Creation, Distribution, Deployment, and Execution -- and identify the structural attack surface each phase introduces. Building on this lifecycle analysis, we construct a threat taxonomy comprising seven categories and seventeen scenarios organized across three attack layers, grounded in both architectural analysis and real-world evidence. We validate the taxonomy through analysis of five confirmed security incidents in the Agent Skills ecosystem. Based on these findings, we discuss defense directions for each threat category, identify open research challenges, and provide actionable recommendations for stakeholders. Our analysis reveals that the most severe threats arise from structural properties of the framework itself, including the absence of a data-instruction boundary, a single-approval persistent trust model, and the lack of mandatory marketplace security review, and cannot be addressed through incremental mitigations alone.

Hyena Operator for Fast Sequential Recommendation

Sequential recommendation models, particularly those based on attention, achieve strong accuracy but incur quadratic complexity, making long user histories prohibitively expensive. Sub-quadratic operators such as Hyena provide efficient alternatives in language modeling, but their potential in recommendation remains underexplored. We argue that Hyena faces challenges in recommendation due to limited representation capacity on sparse, long user sequences. To address these challenges, we propose HyenaRec, a novel sequential recommender that integrates polynomial-based kernel parameterization with gated convolutions. Specifically, we design convolutional kernels using Legendre orthogonal polynomials, which provides a smooth and compact basis for modeling long-term temporal dependencies. A complementary gating mechanism captures fine-grained short-term behavioral bursts, yielding a hybrid architecture that balances global temporal evolution with localized user interests under sparse feedback. This construction enhances expressiveness while scaling linearly with sequence length. Extensive experiments on multiple real-world datasets demonstrate that HyenaRec consistently outperforms Attention-, Recurrent-, and other baselines in ranking accuracy. Moreover, it trains significantly faster (up to 6x speedup), with particularly pronounced advantages on long-sequence scenarios where efficiency is maintained without sacrificing accuracy. These results highlight polynomial-based kernel parameterization as a principled and scalable alternative to attention for sequential recommendation.

Do World Action Models Generalize Better than VLAs? A Robustness Study

Robot action planning in the real world is challenging as it requires not only understanding the current state of the environment but also predicting how it will evolve in response to actions. Vision-language-action (VLA), which repurpose large-scale vision-language models for robot action generation using action experts, have achieved notable success across a variety of robotic tasks. Nevertheless, their performance remains constrained by the scope of their training data, exhibiting limited generalization to unseen scenarios and vulnerability to diverse contextual perturbations. More recently, world models have been revisited as an alternative to VLAs. These models, referred to as world action models (WAMs), are built upon world models that are trained on large corpora of video data to predict future states. With minor adaptations, their latent representation can be decoded into robot actions. It has been suggested that their explicit dynamic prediction capacity, combined with spatiotemporal priors acquired from web-scale video pretraining, enables WAMs to generalize more effectively than VLAs. In this paper, we conduct a comparative study of prominent state-of-the-art VLA policies and recently released WAMs. We evaluate their performance on the LIBERO-Plus and RoboTwin 2.0-Plus benchmarks under various visual and language perturbations. Our results show that WAMs achieve strong robustness, with LingBot-VA reaching 74.2% success rate on RoboTwin 2.0-Plus and Cosmos-Policy achieving 82.2% on LIBERO-Plus. While VLAs such as $π_{0.5}$ can achieve comparable robustness on certain tasks, they typically require extensive training with diverse robotic datasets and varied learning objectives. Hybrid approaches that partially incorporate video-based dynamic learning exhibit intermediate robustness, highlighting the importance of how video priors are integrated.

Online Learnability of Chain-of-Thought Verifiers: Soundness and Completeness Trade-offs

Large language models with chain-of-thought generation have demonstrated great potential for producing complex mathematical proofs. However, their reasoning can often go astray, leading to increasing interest in formal and learned verifiers. A major challenge in learning verifiers, especially when their output will be used by the prover, is that this feedback loop may produce substantial distribution shift. Motivated by this challenge, we propose an online learning framework for learning chain-of-thought verifiers that, given a problem and a sequence of reasoning steps, check the correctness of the solution. Highlighting the asymmetric role of soundness (failure in catching errors in a proof) and completeness (flagging correct proofs as wrong) mistakes of the verifier, we introduce novel extensions of the Littlestone dimension which tightly characterize the mistake bounds for learning a verifier in the realizable setting. We provide optimal algorithms for finding the Pareto-frontier (the smallest total number of mistakes given a budget of soundness mistakes) as well as minimizing a linear combination of asymmetric costs. We further show how our learned verifiers can be used to boost the accuracy of a collection of weak provers, and enable generation of proofs beyond what they were trained on. With the mild assumption that one of the provers can generate the next reasoning step correctly with some minimal probability, we show how to learn a strong prover with small error and abstention rates.

Recursive Models for Long-Horizon Reasoning

Modern language models reason within bounded context, an inherent constraint that poses a fundamental barrier to long-horizon reasoning. We identify recursion as a core principle for overcoming this barrier, and propose recursive models as a minimal realization, where the model can recursively invoke itself to solve subtasks in isolated contexts. We prove that any computable problem admits a recursive decomposition in which each subtask requires only exponentially smaller active context than standard autoregressive models; this strictly surpasses any context management approach confined to a single sequence, such as summarization. We further generalize our framework to modern agentic systems with arbitrary context processing and control flows, and prove that recursive models can achieve optimal power within this broader class. Experimentally, we train a 3B model to reason recursively and evaluate on Boolean satisfiability, a task requiring long-horizon combinatorial search, where it significantly outperforms frontier LLMs.

H-WM: Robotic Task and Motion Planning Guided by Hierarchical World Model

World models are becoming central to robotic planning and control, as they enable prediction of future state transitions. Existing approaches often emphasize video generation or natural language prediction, which are difficult to directly ground in robot actions and suffer from compounding errors over long horizons. Traditional task and motion planning relies on symbolic logic world models, such as planning domains, that are robot-executable and robust for long-horizon reasoning. However, these methods typically operate independently of visual perception, preventing synchronized symbolic and perceptual state prediction. We propose a Hierarchical World Model (H-WM) that jointly predicts logical and visual state transitions within a unified bilevel framework. H-WM combines a high-level logical world model with a low-level visual world model, integrating the robot-executable, long-horizon robustness of symbolic reasoning with perceptual grounding from visual observations. The hierarchical outputs provide stable and consistent intermediate guidance for long-horizon tasks, mitigating error accumulation and enabling robust execution across extended task sequences. To train H-WM, we introduce a robotic dataset that aligns robot motion with symbolic states, actions, and visual observations. Experiments across vision-language-action (VLA) control policies demonstrate the effectiveness and generality of the approach.

SAIL: Self-Amplified Iterative Learning for Diffusion Model Alignment with Minimal Human Feedback

Aligning diffusion models with human preferences remains challenging, particularly when reward models are unavailable or impractical to obtain, and collecting large-scale preference datasets is prohibitively expensive. \textit{This raises a fundamental question: can we achieve effective alignment using only minimal human feedback, without auxiliary reward models, by unlocking the latent capabilities within diffusion models themselves?} In this paper, we propose \textbf{SAIL} (\textbf{S}elf-\textbf{A}mplified \textbf{I}terative \textbf{L}earning), a novel framework that enables diffusion models to act as their own teachers through iterative self-improvement. Starting from a minimal seed set of human-annotated preference pairs, SAIL operates in a closed-loop manner where the model progressively generates diverse samples, self-annotates preferences based on its evolving understanding, and refines itself using this self-augmented dataset. To ensure robust learning and prevent catastrophic forgetting, we introduce a ranked preference mixup strategy that carefully balances exploration with adherence to initial human priors. Extensive experiments demonstrate that SAIL consistently outperforms state-of-the-art methods across multiple benchmarks while using merely 6\% of the preference data required by existing approaches, revealing that diffusion models possess remarkable self-improvement capabilities that, when properly harnessed, can effectively replace both large-scale human annotation and external reward models.

Information-Theoretic Multi-Model Fusion for Target-Oriented Adaptive Sampling in Materials Design

Target-oriented discovery under limited evaluation budgets requires making reliable progress in high-dimensional, heterogeneous design spaces where each new measurement is costly, whether experimental or high-fidelity simulation. We present an information-theoretic framework for target-oriented adaptive sampling that reframes optimization as trajectory discovery: instead of approximating the full response surface, the method maintains and refines a low-entropy information state that concentrates search on target-relevant directions. The approach couples data, model beliefs, and physics/structure priors through dimension-aware information budgeting, adaptive bootstrapped distillation over a heterogeneous surrogate reservoir, and structure-aware candidate manifold analysis with Kalman-inspired multi-model fusion to balance consensus-driven exploitation and disagreement-driven exploration. Evaluated under a single unified protocol without dataset-specific tuning, the framework improves sample efficiency and reliability across 14 single- and multi-objective materials design tasks spanning candidate pools from $600$ to $4 \times 10^6$ and feature dimensions from $10$ to $10^3$, typically reaching top-performing regions within 100 evaluations. Complementary 20-dimensional synthetic benchmarks (Ackley, Rastrigin, Schwefel) further demonstrate robustness to rugged and multimodal landscapes.

Sparsely Supervised Diffusion

Diffusion models have shown remarkable success across a wide range of generative tasks. However, they often suffer from spatially inconsistent generation, arguably due to the inherent locality of their denoising mechanisms. This can yield samples that are locally plausible but globally inconsistent. To mitigate this issue, we propose sparsely supervised learning for diffusion models, a simple yet effective masking strategy that can be implemented with only a few lines of code. Interestingly, the experiments show that it is safe to mask up to 98\% of pixels during diffusion model training. Our method delivers competitive FID scores across experiments and, most importantly, avoids training instability on small datasets. Moreover, the masking strategy reduces memorization and promotes the use of essential contextual information during generation.

AGGC: Adaptive Group Gradient Clipping for Stabilizing Large Language Model Training

To stabilize the training of Large Language Models (LLMs), gradient clipping is a nearly ubiquitous heuristic used to alleviate exploding gradients. However, traditional global norm clipping erroneously presupposes gradient homogeneity across different functional modules, leading to an adverse "spill-over" effect where volatile parameters force unnecessary scaling on stable ones. To overcome this, we propose Adaptive Group-wise Gradient Clipping (AGGC). AGGC partitions parameters into groups based on functional types and regulates each according to its historical behavior using an Exponential Moving Average (EMA). Specifically, it constructs an adaptive interval to simultaneously mitigate gradient explosion and vanishing, while employing a time-dependent scheduling mechanism to balance exploration and convergence. Experiments on LLaMA 2-7B, Mistral-7B, and Gemma-7B models show that AGGC consistently outperforms LoRA and frequently surpasses Full Fine-Tuning. On the GSM8K benchmark, Mistral-7B fine-tuned with AGGC achieves an accuracy of 72.93%, exceeding LoRA's 69.5%. AGGC also effectively stabilizes Reinforcement Learning with Verifiable Rewards (RLVR), enhancing the logic deduction of Qwen 2.5 and Llama 3.2 models. Experimental results demonstrate that AGGC effectively addresses the limitations of traditional gradient clipping methods, particularly in overcoming gradient heterogeneity, by utilizing a modular, adaptive clipping strategy to stabilize the training process. Due to its lightweight design, AGGC can be seamlessly integrated into existing post-training pipelines with negligible overhead.