← Back

Malicious Agent Skills in the Wild: A Large-Scale Security Empirical Study

AI Agent Security arxiv arXiv:2602.06547 PDF β†—
skillswildempiricalscalestudymalicioussecuritymaliciousstudy
Analyzes 98K agent skills from community registries to study the prevalence and nature of malicious third-party agent plugins.
5~10λΆ„. 제λͺ©β†’μ΄ˆλ‘β†’μΈνŠΈλ‘œβ†’μ„Ήμ…˜ν—€λ”β†’κ·Έλ¦Όβ†’κ²°λ‘ λ§Œ.
νŒλ‹¨: μ–΄λ–€ 문제λ₯Ό ν’€κ³  / 핡심 아이디어 / λ‚΄ μž‘μ—…κ³Ό κ΄€λ ¨ μžˆλ‚˜?
~1μ‹œκ°„. κ·Έλ¦ΌΒ·ν‘œλ₯Ό 꼼꼼히. 증λͺ…Β·μˆ˜μ‹ λ””ν…ŒμΌμ€ κ±΄λ„ˆλœ€.
μ‚°μΆœλ¬Ό: "이듀이 뭘 ν–ˆκ³  μ™œ 그게 ν†΅ν•˜λŠ”κ°€" ν•œ 문단.
μž¬ν˜„ν•˜λ“― 읽기. 가정을 μ˜μ‹¬. 직접 인용/λ°˜λ°•ν•  λ…Όλ¬Έλ§Œ.
렌즈: "λ‚΄ ν”Œλ¦Ώμ—μ„œ μΈ‘μ •ν•˜λ©΄ μ €μžκ°€ λͺ» ν•œ 무엇을 보여쀄 수 μžˆλ‚˜?"

Related Papers (10)

Agent Skills in the Wild: An Empirical Study of Security Vulnerabilities at Scale
AI Agent Security unread sim 0.61
CUA-Skill: Develop Skills for Computer Using Agent
Agent Tooling unread sim 0.19
When Single-Agent with Skills Replace Multi-Agent Systems and When They Fail
Multi-Agent unread sim 0.17
AI builds, We Analyze: An Empirical Study of AI-Generated Build Code Quality
Eval & Observability unread sim 0.16
A Large-Scale Study on the Development and Issues of Multi-Agent AI Systems
Multi-Agent unread sim 0.16
MegaFlow: Large-Scale Distributed Orchestration System for the Agentic Era
Agent Tooling unread sim 0.15
Balancing Sustainability And Performance: The Role Of Small-Scale LLMs In Agentic AI Systems
Eval & Observability unread sim 0.15
Corpus2Skill: Don't Retrieve, Navigate β€” Distilling Enterprise Knowledge into Navigable Agent Skills for QA and RAG
Memory & RAG unread sim 0.15
Overcoming the Retrieval Barrier: Indirect Prompt Injection in the Wild for LLM Systems
AI Agent Security unread sim 0.14
Are We All Using Agents the Same Way? An Empirical Study of Core and Peripheral Developers Use of Coding Agents
Eval & Observability unread sim 0.11
View in Knowledge Graph β†’