universal-exam-cram-coach

This module is primarily an orchestrator/test harness with a significant supply-chain/execution risk: it executes an external agent command specified at runtime via --agent-cmd (with the fully constructed prompt substituted into it) using subprocess.Popen, without sandboxing or privilege containment. While the code itself shows no clear embedded malware (no obfuscation payloads, no secret theft, no persistence, no network code), its security posture heavily depends on strict control of --agent-cmd inputs and the environment in which it runs. If an attacker can influence those, they can achieve arbitrary code execution with the runner’s permissions.