Implicit Manipulation for Skill Selection in LLM Agents with Semantic Matching

Abstract

Skill selection is a key stage in LLM-agent workflows, determining which installed skill should handle a user request. Existing attacks on this stage primarily rely on explicit prompt injection or instruction-level steering, which can expose recognizable manipulation signals. In this work, we identify a new implicit attack surface for skill selection: even when the user prompt and skill description appear benign in isolation, their semantic relationship can still be strategically shaped to favor

Categories

Framework mappings

Suggested from the entry's categories.

@misc{wang2026implicit,
  title = {{Implicit Manipulation for Skill Selection in LLM Agents with Semantic Matching}},
  author = {Qikai Wang and Yongzhao Zhang and Zhiwei Chen and Yimiao Sun and Jiguo Yu and Xiaosong Zhang},
  year = {2026},
  month = sep,
  eprint = {2609.02035},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.02035}
}