Modern OS kernels expose hundreds of syscalls to applications, yet individual programs typically require only a small subset. This over-privileging violates the principle of least privilege and expands the kernel attack surface. Existing methods for limiting system calls face several challenges: dynamic tracing suffers from incompleteness, while static analysis, though it ensures completeness, introduces excessive redundant system calls due to a lack of dynamic information and a failure to account for differences in application service scenarios. In this paper, we propose AgentSysFilter, a scenario-aware multi-agent static analysis framework leveraging LLM-driven agents to analyze the service scenarios of the target application and generate scenario-specific syscall limitation strategies. Experimental results indicate that AgentSysFilter reduces allowed syscalls and mitigates more CVEs compared to existing approaches while maintaining functional compatibility. The video and code of AgentSysFilter are available at https://youtu.be/BKWFZL5FR8g and https://github.com/gresces/AgentSysFilter.

@inproceedings{paperbot12610,
  title = {AgentSysFilter: A Multi-agent Analysis Approach for Scenario-Aware Syscall Limitation},
  author = {Dongyang Zhan and Xingchen Yan and Zhaofeng Yu and Yangfan Guo and Kai Tan and Langtong Zhang},
  booktitle = {ASE},
  year = {2026},
  doi = {10.1145/3832783.3834634}
}