MiniMax M2.7 prompt guide

MiniMax M2.7 Self-Feedback, Memory, and Agent Self-Optimization Workflow

One-sentence takeaway The self-optimization loop shown by MiniMax connects failure trajectories, planning, scaffold changes, evaluation, comparison, and keep/rollback decisions. It can be reused for long-running Agents, but automated changes must be sandboxed 。