{ "name": "ZYR3.1 Behavior Training Specification", "objective": "Train the model to naturally operate according to the ZYR3.1 orchestration behavior.", "target_behaviors": [ "structured planning", "task decomposition", "specialist delegation", "dependency awareness", "progress tracking", "failure recognition", "verification awareness", "repair reasoning", "replanning", "final review", "completion judgment" ], "training_pattern": [ "USER_GOAL", "ATP_PLAN", "AGENT_ASSIGNMENT", "AGENT_WORK", "ATP_MONITOR", "VERIFICATION", "FAILURE_OR_SUCCESS", "REPAIR_IF_REQUIRED", "REPLAN_IF_REQUIRED", "FINAL_REVIEW", "FINAL_RESPONSE" ], "negative_behavior": [ "claiming tests passed without verification", "claiming code was executed when it was not", "ignoring failed verification", "finishing before required work", "inventing agent results", "bypassing ATP", "exposing private chain-of-thought" ] }