Cloudflare introduced a tool for automated security auditing

Company Cloudflare released a specialized tool for conducting multi-stage code security checks using AI agents. The solution is focused on automating the process of finding vulnerabilities and forming verifiable reports.

The Security-Audit-Skill system is built around six consecutive phases of work. The process includes reconnaissance, hunting for potential threats, validation of found problems, reporting, and final verification of results. The tool generates machine-readable data, which allows it to be integrated into existing CI/CD pipelines for continuous security monitoring.

A key feature of the development is the use of independent verification at each stage of the audit. This reduces the number of false positives characteristic of standard static code analyzers. The tool is designed to work in conjunction with coding agents that can autonomously perform complex testing scenarios that require a deep understanding of the application context.

The technical implementation of the project is available in open access on GitHub. Developers note that using this approach allows reducing the time for manual security analysis while maintaining a high level of code coverage. Details of the discussion of the architecture and application scenarios are available in the Hacker News branch.

The implementation of such automated skills for AI agents changes the approach to development security. Instead of periodic manual checks, companies have the opportunity to conduct a deep audit of each code iteration, which is critical for protecting infrastructure from modern attack vectors.