StepSecurity adds inventories to track AI agent plugins

StepSecurity has introduced two complementary inventories to map the spread and scope of AI agent plugins for Claude Code and Codex. Dev Machine Guard now inventories plugins installed on developers’ machines, capturing which plugins exist, where they came from (marketplace, GitHub catalog, local dir, etc.), their versions, and whether they are enabled. A separate Agent Plugins page under GitHub>AI Agents shows plugins published or enabled in an organisation’s repositories.

Together these views help security teams answer what agents have been extended with on developers’ machines and what repositories are distributing to the organisation, the company explains in their blog post dated 26 October 2026.

The article emphasises that agent plugins constitute a supply chain surface because a plugin is a bundle capable of running code, declaring identity, and being distributed at scale. It notes three recurring risk factors: plugins can execute code on a developer’s machine with user privileges; plugin identity is self-declared in manifests; and repositories can publish or enable plugins for broad deployment across teams.

StepSecurity then describes how the two inventories operate: Dev Machine Guard scans and reports plugin metadata without executing code or transmitting sensitive data, while the GitHub view lists plugin name, author, manifest details, repository path, version, and which repositories have the plugin enabled.

The piece also references prior attack patterns that could exploit such surfaces—credential-harvesting hooks in a .claude/settings.json file, fake AI tools on trusted marketplaces, and automated pushes from compromised maintainers—illustrating why cross-referencing both views is important for tracing provenance and exposure.