Large language model coding agents, such as Claude Code, are being applied to security auditing. A new preprint on arXiv (2610.25705) examines how effectively these agents detect vulnerabilities in the Linux kernel when equipped with task-specific security skills. Such skills are meant to encode reusable procedures for vulnerability analysis.
The abstract outlines the motivation: coding agents are increasingly used for security auditing, and skills can provide structured guidance. However, the abstract cuts off before presenting findings, so the reported effectiveness remains unclear from the announcement alone. The study appears to be an evaluation of this approach in the context of the Linux kernel, a critical open-source codebase.
Given the limited information in the abstract, the article can only note the research direction. The preprint's full results would be needed to assess whether security skills meaningfully improve detection rates.