Query-Only Backdoor Attacks on Self-Evolving Skills via Trajectory Poisoning

Abstract

Agentic skills improve large language model (LLM) agents by encoding reusable procedures for complex tasks. However, manually authored skills often adapt poorly to long-horizon tasks and changing environments. To address the limitation, self-evolving skill systems have been developed to automatically construct and update skills from execution trajectories, shifting skill acquisition from external marketplaces to a trusted evolution pipeline. By replacing external skill acquisition with trusted i

Categories

Framework mappings

Suggested from the entry's categories.

@misc{luo2026queryonly,
  title = {{Query-Only Backdoor Attacks on Self-Evolving Skills via Trajectory Poisoning}},
  author = {Yuyang Luo and Haoran Wang and Kai Shu},
  year = {2026},
  month = aug,
  eprint = {2608.08303},
  archivePrefix = {arXiv},
  url = {https://www.semanticscholar.org/paper/b4119fc7267c95cfc9c974be181981e520e73a64}
}