Preliminary research

Chaining Skills to Hijack LLM Agents

Tian Dong, Zixuan Ma, Haodong Zhao, Huaien Zhang, Shaofeng Li and Hao Chen · arXiv · 1 October 2026

AI-collected research leads through 3 November 2026, including a November–October–September missed-article review and a targeted multilingual search alongside earlier community-source sweeps. Unranked, incomplete, not community-vetted, and subject to change.

APEX constructs adversarial skill chains that mix real task progress with false approval claims in persistent handoff records. It evaluates executed target actions, original-task checks, repeated-work costs, and a request-checking prompting defence.

Record

Researcher

Tian Dong, Zixuan Ma, Haodong Zhao, Huaien Zhang, Shaofeng Li and Hao Chen

Published by

arXiv

Date

1 October 2026

In the archive

• Collected in the 2026 preliminary research sweep

Related sources

• Code

Tags

• ai-agent
• skills
• prompt-injection
• access-control
• supply-chain
• owasp-a03-2021
• owasp-a06-2021

This page is the archive's own catalogue record. The research is the work of Tian Dong, Zixuan Ma, Haodong Zhao, Huaien Zhang, Shaofeng Li and Hao Chen, first published at the original source. Preserved copies are kept so the citation survives its host.