New Plugin Enhances Governance for AI Code Assistants
JFrog has launched a new plugin designed to provide developers with governed access for analyzing, validating, and securing each artifact and dependency consumed by their coding agents. This tool governs the components an agent interacts with within the software supply chain, without altering the agent’s behavior. This initiative reflects the explosive increase in binaries generated through AI-assisted development, with JFrog reporting over 18 billion artifacts managed on its platform—a staggering 136% year-over-year increase.
The Israeli-American software company positions its platform as a trusted layer and a reference system for agent-based coding environments. The new plugin complements JFrog’s recently announced MCP Registry and Agent Skills Registry, with plans to extend support to other coding agents, starting with Cursor and VS Code Copilot.
This announcement comes at a time when coding agents are fundamentally altering the software supply chain. These agents often make decisions regarding dependencies, builds, and deployments autonomously, frequently lacking adequate visibility into their environments and contextual knowledge of the supply chain. This can lead to the unintentional introduction of malicious packages, vulnerabilities, and uncontrolled AI resources into production, putting organizations at risk of supply chain attacks.
The scale of this phenomenon underscores the urgent need for governance. The rise of code production at scale by AI assistants has contributed to a resurgence of classic vulnerabilities. In 2025 alone, more than 48,000 new vulnerabilities were reported, a 20% increase compared to the previous year. JFrog’s 136% growth in managed artifacts mirrors this dynamic, reflecting a software supply chain that evolves faster than the governance frameworks set in place to manage it.
Governing Agent Consumption
The new plugin embeds governance, security, license compliance, and provenance validation directly into the development flow, rather than addressing these issues post-factum. Policies are enforced during code writing, eliminating the need for manual transfers that can delay production releases. Traceability extends from source commits to build artifacts, a crucial aspect for security teams that must respond swiftly to audits or incidents.
It’s important for security professionals to note that while the plugin controls the artifacts and dependencies consumed by the agent, it does not alter the agent’s reasoning or ensure its behavior is safe. The governance focuses on the software supply chain—where compromised components can infiltrate—leaving the question of the intrinsic reliability of the agent itself untouched. This integration of AI artifacts as application components precedes governance frameworks, which still struggle to keep pace.
Governance of MCP Servers and Agent Skills
The plugin restricts usage to verified MCP servers and agent capabilities, thereby preventing unauthorized access to sensitive internal data. Additionally, the platform offers three connectivity levels designed to operate in any coding environment. JFrog Platform Skills equip agents with specific knowledge for tasks such as vulnerability analysis and provenance checks, while JFrog MCP Tools provide standardized access to security, compliance, and artifact data.
However, the scope of this control also defines its limitations. Governance only applies to artifacts, dependencies, and MCP servers that traverse the JFrog platform. An agent retrieving a dependency outside the registry or connecting to an ungoverned MCP server will not be subject to these controls. Consequently, decision-makers must channel all agent activities through the platform to ensure effective governance coverage.
JFrog as a Reference System for Multi-Agent Environments
JFrog believes that teams will increasingly employ multiple agents, driving value towards a cross-cutting governance layer that is independent of the employed agent. This ambition positions the platform at the core of continuous integration and delivery processes, providing auditability and provenance, while introducing an additional control point and dependency that each organization must weigh.
To support this shift, Anthropic has called for a collective investment in a security posture focused on agents, advocating for shared assessment repositories, common identity standards, and red teaming exercises across suppliers. This forward-looking approach transcends the plugin’s scope, which remains a building block for dependency governance. As the software supply chain emerges as a primary attack vector, IT departments must track the origin of every component, especially as the forthcoming AI Act and enhanced European cybersecurity mandates bolster traceability requirements.
For IT leaders industrializing AI-assisted development, the evaluation isn’t whether to implement governance but rather about the real extent of coverage. A plugin that secures agent-consumed dependencies mitigates a known portion of risk, provided all traffic is channeled through the platform, without assuming the agent’s reliability. Securing coding agents remains an ongoing industry challenge, of which this integration is a crucial component.



