agent-bom runtime

This Skill addresses the lack of security monitoring for AI agents during runtime. In real-world deployments, AI agents may generate anomalous context call chains, lateral movement behaviors, or exploit unknown vulnerabilities, which traditional monitoring tools struggle to cover. agent-bom runtime helps teams quickly identify security risks by constructing context graphs and correlating audit logs with CVE discoveries.

Usage is straightforward. After installation via pipx, you can invoke three core tools using natural language commands. For example, execute context_graph() to generate a context relationship diagram for the current AI agent, use runtime_correlate() to cross-reference runtime audit logs with known CVEs, or query vulnerability trends and runtime event history via analytics_query(). All operations are performed in-memory, requiring no network connection.

It is well-suited for operations teams and agent developers who need to ensure the security of AI agents. This is particularly valuable for enterprises that have deployed multiple AI coding agents or automated workflows and require continuous monitoring of agent behavior, audit logging, and investigation of potential vulnerabilities. Additionally, security auditors can leverage it to quickly generate compliance reports.

We recommend integrating this tool into your CI/CD pipeline to perform a runtime security scan before each deployment. Note that it primarily relies on a Python 3.11+ environment, with ClickHouse available as an optional dependency for persistent analysis storage. It does not automatically discover files or initiate network requests; users must manually provide audit log files.