An injection vulnerability has been discovered in the Skills Guard Multi‑Word Prompt Handler component of NousResearch hermes‑agent. By manipulating the THREAT_PATTERNS argument in agent/skills_guard.py, an attacker can supply crafted input that is executed by the system. This flaw is a form of code injection that can lead to remote code execution. The weakness aligns with CWE‑74 for improper filtering and CWE‑707 for XML external entity mismanagement.

This issue affects all installations of NousResearch hermes‑agent up to version 2026.4.23, as disclosed in the vendor product notes. No specific sub‑components or library versions are mentioned beyond the main module. Systems running newer releases are presumed unaffected but should verify the version.

The CVSS score of 6.9 indicates a medium‑to‑high severity, while the EPSS score is not provided, suggesting the exploitation likelihood is not quantified. Because the vulnerability allows remote exploitation and the exploit has been publicly disclosed, the risk is significant for exposed deployments. As the feature is not listed in CISA KEV, the criticality is moderate, yet administrators should treat it as a high‑priority patch once a fix is released.