A flaw in the _serve_plugin_skill/skill_view function located in tools/skills_tool.py of NousResearch Hermes Agent allows an attacker to manipulate input data in a way that causes injection of arbitrary code or commands. This vulnerability can be leveraged from a remote location, giving the attacker the ability to execute unintended instructions within the agent’s runtime environment. The resulting impact is the potential compromise of system confidentiality, integrity, or availability, as the injected payload could modify agent behavior or exfiltrate sensitive data.

Any deployment of NousResearch Hermes Agent up to and including version 2026.4.30 is potentially affected. The vulnerability is not present in later releases, but without a confirmed patch timeline the only known mitigation is to upgrade beyond the stated version or otherwise remove the exposed function.

The CVSS base score of 6.9 reflects a moderate severity, combined with a publicly disclosed exploit that may be available in the wild. Although the EPSS score is not published, the lack of a KEV listing suggests the vulnerability has not yet escalated to a high priority attack surface, yet the remote injection capability warrants prompt attention. Attackers would need to reach the agent’s endpoint and submit crafted input; if successful, they could gain code execution privileges on the host running the agent.