MMSkillRisk: Can Agents Stay Safe When Multimodal Skills Become Traps?

arXiv:2609.35912 · cs.CR, cs.AI · Submitted 2026-09-28 · Read on arXiv

cs.CR, cs.AI

Submitted: 2026-09-28

Updated: 2026-09-28

Code: https://github.com/kaill-jlq/MMSkillRisk

Terminology

• Trace — 15× in this paper · explained in 2 episodes
• Agent Skills — 12× in this paper · explained in 1 episode
• Benchmark — 12× in this paper · explained in 2 episodes
• Attack Success Rate — 4× in this paper
• Native-Context Visual Attack — 4× in this paper
• Ablation — 3× in this paper · explained in 1 episode
• Adaptation — 3× in this paper · explained in 1 episode
• Task Success Rate — 3× in this paper
• Verification — 3× in this paper · explained in 2 episodes
• Ablation Studies — 2× in this paper · explained in 1 episode
• Aggregation — 2× in this paper · explained in 1 episode
• Benchmarking — 2× in this paper · explained in 3 episodes
• Grounding — 2× in this paper · explained in 3 episodes
• LLM judge — 2× in this paper · explained in 2 episodes
• Visual Instruction Adoption Rate — 2× in this paper
• Actor — 1× in this paper · explained in 3 episodes
• Agentic AI — 1× in this paper · explained in 1 episode
• Attack Success Rate (ASR) — 1× in this paper · explained in 2 episodes
• Continual Learning — 1× in this paper · explained in 2 episodes
• Core — 1× in this paper · explained in 2 episodes
• Embeddings — 1× in this paper · explained in 2 episodes
• KV cache — 1× in this paper · explained in 3 episodes
• Multi-agent Collaboration — 1× in this paper · explained in 2 episodes
• Procedural Knowledge — 1× in this paper · explained in 2 episodes
• Scaffolding — 1× in this paper · explained in 3 episodes
• Visual Grounding — 1× in this paper · explained in 2 episodes

Sources

• Formal Analysis and Supply Chain Security for Agentic AI Skills
• MMG2Skill: Can Agents Distill In-the-Wild Guides into Self-Evolving Skills?
• DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression
• RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources
• MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills
• SkillProbe: Security Auditing for Emerging Agent Skill Marketplaces via Multi-Agent Collaboration
• SkillJect: Effectively Automating Skill-Based Prompt Injection for Skill-Enabled Agents
• Seeing Is Not Screening: Multimodal Hidden Instruction Attacks on Agent Skill Scanners
• XSkill: Continual Learning from Experience and Skills in Multimodal Agents
• HarmfulSkillBench: How Do Harmful Skills Weaponize Your Agents?
• VISUALSKILL: Multimodal Skills for Computer-Use Agents
• SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces
• Kimi K3: Open Frontier Intelligence
• SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks
• OpenSkillRisk: Benchmarking Agent Safety When Using Real-World Risky Third-Party Skills
• Agent Skills in the Wild: An Empirical Study of Security Vulnerabilities at Scale
• Supply-Chain Poisoning Attacks Against LLM Coding Agent Skill Ecosystems
• On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability
• Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks
• Voyager: An Open-Ended Embodied Agent with Large Language Models

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills
• Token Composition: A Graph Based on EVM Logs