auto-review-loop-minimax

SUSPICIOUS. The MiniMax data flow itself is coherent and direct-to-official endpoints, but the skill’s autonomy is disproportionate: it can iteratively modify code, run experiments, use SSH on remote servers, and explicitly tells the agent to perform some actions silently without user approval. This is not credential harvesting, but it is a high-risk autonomous workflow skill.