CVE-2026-51866 describes a vulnerability in DB-GPT versions 0.7.5 and 0.8.0 where a skill uploaded through the legitimate /api/v1/skills/upload route can later be executed via the /api/v1/chat/react-agent route. The vulnerability involves the interaction between these two API endpoints allowing execution of uploaded skills, potentially leading to unauthorized or unintended execution of code or actions within the application context.

The impact is that an uploaded skill can be executed through a different API endpoint, which may allow an attacker or user to trigger functionality that was not intended to be executed in that manner. The exact consequences depend on the nature of the uploaded skill and the permissions of the executing context. No known exploits in the wild have been reported.

No patch or official fix information is provided. Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, restrict access to the upload and react-agent endpoints to trusted users only and monitor usage for suspicious activity.