Behavioral Skill Reconstruction: Reconstructing Hidden Functionality from LLM Agent Skills
cs.CR, cs.AI, cs.CL
Submitted: 2026-08-04
Updated: 2026-09-19
Comments: 20 pages, 5 figures, 20 tables
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- SkillRet: A Large-Scale Benchmark for Skill Retrieval in LLM Agents
- Agent Harness Distillation: Inference-Time Harness Extraction and Exploitation in Autonomous Multi-Agent Systems
- Spore: Efficient and Training-Free Privacy Extraction Attack on LLMs via Inference-Time Hybrid Probing
- Let Them Steal: Trapping Large Language Model Extraction Attacks with Knowledge Honeypot
- Agent Skills Matter: Inferring Proprietary Skills from Execution Trajectories
- Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection
- Distilling the Knowledge in a Neural Network
- SkillVetBench: LLM-as-Judge for Multi-Dimensional Security Risk Evaluation in Open-Source LLM Agent Skills
- SkillMutator: Benchmarking and Defending Language-and-Code Cross-modal Attacks on LLM Agent Skills
- Runtime Skill Audit: Targeted Runtime Probing for Agent Skill Security
- SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks
- SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution
- Agent Skills in the Wild: An Empirical Study of Security Vulnerabilities at Scale
- Ignore Previous Prompt: Attack Techniques For Language Models
- Prompt Stealing Attacks Against Large Language Models
- Kimi K2: Open Agentic Intelligence
- Voyager: An Open-Ended Embodied Agent with Large Language Models
- Black-Box Skill Stealing Attack from Proprietary LLM Agents: An Empirical Study
- Agent Workflow Memory
- Benign in Isolation, Harmful in Composition: Security Risks in Agent Skill Ecosystems
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
- Estimating the Social Cost of Corporate Data Breaches
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills