Skill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning

Agent skills are callable procedural modules that provide reusable knowledge and execution policies for complex agentic tasks. However, existing methods mainly focus on selecting relevant skills or improving the skills themselves, while overlooking whether a relevant skill should actually be invoked at the current decision point. Unhelpful invocations may introduce irrelevant context and disrupt an otherwise correct execution process. To address this issue, we propose SelSkill, a dual-granularity preference-learning framework for selective skill invocation. SelSkill formulates skill use as a skill-or-skip decision, uses predictive uncertainty to prioritize candidate decision points, and constructs controlled invoke-skip preference pairs from shared trajectory prefixes. It further combines episode-level outcome preferences with step-level invocation preferences to capture both overall trajectory quality and the local effectiveness of skill invocation. On ALFWorld with Qwen3-8B, SelSkill improves task success by 10.9 percentage points and execution precision by 29.1 percentage points. On BFCL, it improves task success by 5.7 percentage points and execution precision by 29.5 percentage points. Zero-shot results on Tau-bench and PopQA further suggest that the learned invocation policy transfers to new domains with previously unseen skills.

EvoStreaming: Your Offline Video Model Is a Natively Streaming Assistant

Streaming video understanding demands more than watching longer videos: assistants must decide when to speak in real time, balancing responsiveness against verbosity. Yet most video-language models (VideoLLMs) are trained for offline inference, and existing streaming benchmarks externalize this timing decision to the evaluator. We address this gap with RealStreamEval, a frame-level multi-turn evaluation protocol that exposes models to sequential observations and penalizes unnecessary responses. Under this protocol, we observed that strong offline VideoLLMs retain useful visual understanding but lack an interaction policy for deciding when to respond. Motivated by this observation, we propose EvoStreaming, a self-evolved streaming adaptation framework in which the base model itself acts as data generator, relevance annotator, and roll-out policy to synthesize streaming trajectories without external supervision. With only $1{,}000$ self-generated samples ($139\times$ less than the leading streaming instruction-tuning approach) and no architectural changes, EvoStreaming consistently improves the overall RealStreamEval score by up to $10.8$ points across five open VideoLLM backbones (Qwen2/2.5/3-VL, InternVL-3.5, MiniCPM-V4.5) while largely preserving offline video performance. These results suggest that data-efficient interaction tuning is a practical path for adapting existing VideoLLMs to streaming assistants.

StreamMeCo: Long-Term Agent Memory Compression for Efficient Streaming Video Understanding

Vision agent memory has shown remarkable effectiveness in streaming video understanding. However, storing such memory for videos incurs substantial memory overhead, leading to high costs in both storage and computation. To address this issue, we propose StreamMeCo, an efficient Stream Agent Memory Compression framework. Specifically, based on the connectivity of the memory graph, StreamMeCo introduces edge-free minmax sampling for the isolated nodes and an edge-aware weight pruning for connected nodes, evicting the redundant memory nodes while maintaining the accuracy. In addition, we introduce a time-decay memory retrieval mechanism to further eliminate the performance degradation caused by memory compression. Extensive experiments on three challenging benchmark datasets (M3-Bench-robot, M3-Bench-web and Video-MME-Long) demonstrate that under 70% memory graph compression, StreamMeCo achieves a 1.87* speedup in memory retrieval while delivering an average accuracy improvement of 1.0%. Our code is available at https://github.com/Celina-love-sweet/StreamMeCo.

Consistency-aware Fake Videos Detection on Short Video Platforms

This paper focuses to detect the fake news on the short video platforms. While significant research efforts have been devoted to this task with notable progress in recent years, current detection accuracy remains suboptimal due to the rapid evolution of content manipulation and generation technologies. Existing approaches typically employ a cross-modal fusion strategy that directly combines raw video data with metadata inputs before applying a classification layer. However, our empirical observations reveal a critical oversight: manipulated content frequently exhibits inter-modal inconsistencies that could serve as valuable discriminative features, yet remain underutilized in contemporary detection frameworks. Motivated by this insight, we propose a novel detection paradigm that explicitly identifies and leverages cross-modal contradictions as discriminative cues. Our approach consists of two core modules: Cross-modal Consistency Learning (CMCL) and Multi-modal Collaborative Diagnosis (MMCD). CMCL includes Pseudo-label Generation (PLG) and Cross-modal Consistency Diagnosis (CMCD). In PLG, a Multimodal Large Language Model is used to generate pseudo-labels for evaluating cross-modal semantic consistency. Then, CMCD extracts [CLS] tokens and computes cosine loss to quantify cross-modal inconsistencies. MMCD further integrates multimodal features through Multimodal Feature Fusion (MFF) and Probability Scores Fusion (PSF). MFF employs a co-attention mechanism to enhance semantic interactions across different modalities, while a Transformer is utilized for comprehensive feature fusion. Meanwhile, PSF further integrates the fake news probability scores obtained in the previous step. Extensive experiments on established benchmarks (FakeSV and FakeTT) demonstrate our model exhibits outstanding performance in Fake videos detection.