LLM tool-use agents operate in dynamic environments where many actions carry operational risk. However, most safety mechanisms react only after errors manifest. Existing pre-emptive approaches either fine-tune the agent on chain-of-thought deliberation or compile natural-language guardrails into runtime checks, but they do so without exposing a structural, auditable verdict. We propose POLAR, a guardrail framework for small tool-calling agents that assesses reversibility through a structured two-layer ontology. POLAR assigns each action a graded reversibility score by deriving a candidate inverse sequence; calls failing a threshold are pruned before execution. Evaluated on 𝜏²-bench across six agent models, POLAR improves mean task reward by 0.11 to 0.18 points on airline for four of six agents, but only eight of eighteen model–domain cells improve overall; retail and stronger agents often regress. POLAR provides an auditable structural check and characterizes its task-utility trade-offs. Reward is not a direct measure of prevented harm.

LLM-based agent systems increasingly adopt skill-based architectures to reduce repetitive reasoning costs and improve stable, efficient task execution. Recent studies propose self-evolving agents that autonomously generate, refine, and reuse skills from past experiences to enable continuous capability evolution. However, autonomous skill evolution introduces a new attack surface in which malicious capabilities are generated, stored, and reused as legitimate skills. In this paper, we define EvoSkill Injection as a threat model targeting the autonomous skill generation and evolution pipeline of self-evolving agents. We further propose SARGE (Red-teaming Autonomous Skill Generation and Evolution in self-evolving agents), a red-teaming framework for evaluating this threat model through iterative generation, escalation, and reinforcement interactions. To support our framework, we construct EvoSkillBench, a benchmark dataset of malicious interaction trajectories for inducing malicious skill formation in self-evolving agents, and introduce EvoSkillSafetyBench to assess whether injected malicious skills are retrieved and activated as harmful behaviors during downstream task execution. Our evaluation shows that SARGE induces malicious skill formation and that injected skills are persistently stored and repeatedly activated, highlighting the risk of persistent capability corruption.

Multimodal emotion recognition (MER) in conversation is inherently ambiguous, as linguistic, acoustic, and visual cues may support multiple plausible interpretations. Existing approaches often treat ambiguity as prediction uncertainty rather than modeling its structured sources. We propose UAF, an ambiguity-aware framework for MER built around a single unified ambiguity signal. UAF estimates a normalized per-sample Model-based Ambiguity Degree (MAD) from the predicted emotion distribution by measuring its proximity to the maximally ambiguous uniform distribution, and uses it as a shared control signal rather than a post-hoc uncertainty score. This single scalar coordinates three parts of learning: it aligns representation radius with ambiguity, emphasizes ambiguous samples during optimization, and scales OT-based perturbation consistency. The fused evidence is organized into context, speaker, temporal, cross-modal, and residual subspaces that serve as the substrate on which this signal operates. Experiments on IEMOCAP, CMU-MOSEI, and MELD show that UAF yields consistently better calibration and high-ambiguity reliability alongside competitive accuracy and weighted-F1, showing that a single ambiguity signal can be an effective shared training signal for reliable MER