The vulnerability arises from the AgentService.loadAgentFromFile method parsing YAML files with the js-yaml library without disabling dangerous tags such as !!js/function and !!js/undefined. This oversight permits an attacker to craft a malicious YAML file that, when parsed by the server, causes arbitrary JavaScript code to execute. The result is a remote code execution vulnerability that exposes the server to full compromise. Specifically, the weakness is categorized as improper deserialization of untrusted data (CWE-502).

The affected product is PraisonAI, a multi‑agent teams system developed by MervinPraison. All releases prior to 4.5.115 are vulnerable. Versions 4.5.115 and later contain the fix and are not at risk.

The CVSS score of 9.8 indicates a high severity level, with nearly maximum potential impact on confidentiality, integrity, and availability. The EPSS score is not available, and the vulnerability is not currently listed in CISA’s KEV catalog, suggesting it has not yet been widely exploited. The likely attack vector is the API endpoint that accepts agent definition uploads; an attacker who can submit a malicious YAML file through this interface can trigger the vulnerability and obtain remote code execution on the server.