The vulnerability resides in the _scan_context_content function of the Hermès Agent’s prompt_builder.py module. A manipulated request can trigger an injection payload that is executed by the agent. This flaw allows an attacker to inject arbitrary commands or code and gain control over the execution environment, potentially compromising confidentiality, integrity, and availability of the target system. The description indicates that the attack vector is remote and that the exploit has already been published.

Vendors affected include NousResearch, specifically the Hermès Agent product version 2026.4.23. The CPE identifier for the affected component is cpe:2.3:a:nousresearch:hermes-agent:*:*:*:*:*:*:*. No other affected versions or products were identified in the current data set.

The flaw carries a CVSS score of 6.9, reflecting a moderate severity. The EPSS score is not available, and the vulnerability is not listed in the CISA KEV catalog, suggesting it has not yet been widely exploited in the wild. However, because the exploit code is publicly available and the attack can be performed from a remote location, the risk to exposed systems remains significant. An attacker would need network access to the agent’s exposed endpoint, but no local privilege escalation or authentication is required according to the description.