A command injection vulnerability in Evolver, a GEP-powered self-evolving engine for AI agents, affects versions prior to 1.69.3. The vulnerability exists in the _extractLLM() function which constructs curl commands using string concatenation without proper sanitization. Attackers can execute arbitrary shell commands on the server by injecting shell metacharacters into the corpus parameter. The function passes unsanitized input to execSync(), enabling remote code execution. This critical security flaw has been patched in version 1.69.3.